20130327 Sharding Pinterest

合集下载

pinterest的简介,及具体操作16页

Pinterest中文名字：拼趣
中文名诠释：拼之趣图，快乐分享！
Pinterest是一个创意组合词，由Pin和Interest组合而成，音标读作 [ˈPintrist]，
原义是指把自己喜欢的图片就像拿钉子钉在白板上一样分享。Pinterest 中文名字中叫的最好听的是“拼趣”。“拼趣”一词属于创造性的翻译，既保留了原英文品牌的大部分发音，Pinterest名称由Pin（拼） +Interest（兴趣）组成，寓意为把自己感兴趣的东西(图片)用图钉钉在钉板（PinBoard）上，让用户不断发现新图片，又不拘泥于刻板的音译手法，该名字有“拼之趣图”的另一层深刻的含义，而且朗朗上口，含义丰富，属于上等高雅之名字。
2019年5月20日，美国图片社交网站Pinterest即将推出三类新“图钉”(pin)，迈出与美国大型零售品牌合作的第一步
操ቤተ መጻሕፍቲ ባይዱ步骤
• 转载分享 like
• 把鼠标移动那就会出现如左图的图标
• 本文由瑞诺国际(外贸建站外贸营销) reanod提供 •
谢谢！
Pin：把你喜欢的图片 Pin 到一块白板（Board）上，类似于做简报。
Repin：图钉一旦被添加，就可以被其他Pinterest用户转钉。这是网站内容得以病毒式传播的原理。如果你在P发现喜欢的东西，可以通过转钉来与朋友分享。
Board：存放所有图钉的地方，可以针对不同主题设置独立面板。相当于一本独立的相册，按照不同的主题和兴趣来划分。
2019年4月，合创人之一保罗·夏拉离开Pinterest。 2019年1月24日，Pinterest获得风险投资公司SV Angel 3000万美元的
风险投资。此次交易，Pinterest的估值为15亿美元。

pinterest广告投放模式及实操技巧

pinterest广告投放模式及实操技巧
Pinterest广告投放模式是基于用户兴趣和行为进行定向广告投放的。

下面是在Pinterest进行广告投放的实操技巧：
1. 目标受众定位：使用Pinterest提供的广告管理工具，根据目标受众的兴趣、行为、地理位置等进行定向投放。

Pinterest有丰富的用户数据，可以根据用户在平台上的搜索和浏览行为来确定您的目标受众。

2. 广告格式选择：Pinterest广告提供了多种不同的广告格式，包括标准推广、购物推广、品牌推广和应用推广等。

根据您的广告目标选择适合的广告格式。

3. 制作优质图片和精选标题：在Pinterest上，图片是最重要的元素之一。

制作高品质的图片和吸引人的标题，可以吸引更多的用户点击和互动。

4. 使用关键词进行定向：Pinterest使用关键词匹配来为广告进行定向。

通过研究和理解您的目标受众使用的关键词，可以选择适当的关键词进行定向投放。

5. 设置合适的投放预算和投放时间：根据您的广告目标和资源限制，设定合适的广告投放预算和投放时间。

可以通过测试不同预算和投放时间来优化广告的效果。

6. 监测和优化广告效果：定期监测广告的点击率、转化率和互动率等关键指标，根据数据进行优化和调整。

可以尝试不同的
广告创意、标题和定向等来提高广告的效果。

最重要的是，了解您的目标受众，理解Pinterest用户的行为和兴趣，才能更好地进行广告投放。

XC2C32A-6QFG32I中文资料

© 2004-2007 Xilinx, Inc. All rights reserved. All Xilinx trademarks, registered trademarks, patents, and disclaimers are as listed at /legal.htm .All other trademarks and registered trademarks are the property of their respective owners. All specifications are subject to change without notice.Features•Optimized for 1.8V systems-As fast as 3.8ns pin-to-pin logic delays -As low as 12 μA quiescent current •Industry’s best 0.18 micron CMOS CPLD-Optimized architecture for effective logic synthesis -Multi-voltage I/O operation: 1.5V through 3.3V •Available in multiple package options -32-land QFN with 21 user I/O -44-pin PLCC with 33 user I/O -44-pin VQFP with 33 user I/O -56-ball CP BGA with 33 user I/O -Pb-free available for all packages •Advanced system features-Fastest in system programming· 1.8V ISP using IEEE 1532 (JTAG) interface -IEEE1149.1 JTAG Boundary Scan Test -Optional Schmitt-trigger input (per pin)-Two separate I/O banks-RealDigital 100% CMOS product term generation -Flexible clocking modes-Optional DualEDGE triggered registers -Global signal options with macrocell control·Multiple global clocks with phase selection permacrocell ·Multiple global output enables ·Global set/reset-Efficient control term clocks, output enables andset/resets for each macrocell and shared across function blocks-Advanced design security-Open-drain output option for Wired-OR and LEDdrive-Optional configurable grounds on unused I/Os -Optional bus-hold, 3-state or weak pullup onselected I/O pins-Mixed I/O voltages compatible with 1.5V, 1.8V,2.5V, and3.3V logic levels -PLA architecture·Superior pinout retention ·100% product term routability across functionblock-Hot pluggableRefer to the CoolRunner™-II family data sheet for architec-ture description.DescriptionThe CoolRunner ™-II 32-macrocell device is designed for both high performance and low power applications. This lends power savings to high-end communication equipment and high speed to battery operated devices. Due to the low power stand-by and dynamic operation, overall system reli-ability is improvedThis device consists of two Function Blocks interconnected by a low power Advanced Interconnect Matrix (AIM). The AIM feeds 40 true and complement inputs to each Function Block. The Function Blocks consist of a 40 by 56 P-term PLA and 16 macrocells which contain numerous configura-tion bits that allow for combinational or registered modes of operation.Additionally, these registers can be globally reset or preset and configured as a D or T flip-flop or as a D latch. There are also multiple clock signals, both global and local product term types, configured on a per macrocell basis. Output pin configurations include slew rate limit, bus hold, pull-up,open drain and programmable grounds. A Schmitt trigger input is available on a per input pin basis. In addition to stor-ing macrocell output states, the macrocell registers may be configured as "direct input" registers to store signals directly from input pins.Clocking is available on a global or Function Block basis.Three global clocks are available for all Function Blocks as a synchronous clock source. Macrocell registers can be individually configured to power up to the zero or one state.A global set/reset control line is also available to asynchro-nously set or reset selected registers during operation.Additional local clock, synchronous clock-enable, asynchro-nous set/reset and output enable signals can be formed using product terms on a per-macrocell or per-Function Block basis.The CoolRunner-II 32-macrocell CPLD is I/O compatible with standard LVTTL and LVCMOS18, LVCMOS25, and LVCMOS33 (see Table 1). This device is also 1.5V I/O com-patible with the use of Schmitt-trigger inputs.Another feature that eases voltage translation is I/O bank-ing. Two I/O banks are available on the CoolRunner-II 32A macrocell device that permit easy interfacing to 3.3V, 2.5V,1.8V, and 1.5V devices.XC2C32A CoolRunner-II CPLDDS310 (v2.0) March 8, 2007Product Specification2DS310 (v2.0) March 8, 2007RealDigital Design TechnologyXilinx CoolRunner-II CPLDs are fabricated on a 0.18 micron process technology which is derived from leading edge FPGA product development. CoolRunner-II CPLDs employ RealDigital, a design technique that makes use of CMOS technology in both the fabrication and design methodology.RealDigital design technology employs a cascade of CMOS gates to implement sum of products instead of traditional sense amplifier methodology. Due to this technology, Xilinx CoolRunner-II CPLDs achieve both high performance and low power operation.Supported I/O StandardsThe CoolRunner-II 32 macrocell features both LVCMOS and LVTTL I/O implementations. See Table 1 for I/O stan-dard voltages. The LVTTL I/O standard is a general purpose EIA/JEDEC standard for 3.3V applications that use an LVTTL input buffer and Push-Pull output buffer. TheLVCMOS standard is used in 3.3V, 2.5V, 1.8V applications.CoolRunner-II CPLDs are also 1.5V I/O compatible with the use of Schmitt-trigger inputs.Table 1: I/O Standards for XC2C32A IOSTANDARD Attribute Output V CCIO Input V CCIO Input V REF BoardTermination Voltage V T LVTTL 3.3 3.3N/A N/A LVCMOS33 3.3 3.3N/A N/A LVCMOS25 2.5 2.5N/A N/A LVCMOS18 1.8 1.8N/A N/A LVCMOS15(1)1.51.5N/AN/A(1) LVCMOS15 requires Schmitt-trigger inputs.Figure 1: I CC vs FrequencyTable 2: I CC vs Frequency (LVCMOS 1.8V T A = 25°C)(1)Frequency (MHz)255075100150175200225250300Typical I CC (mA)0.0160.871.752.613.445.165.996.817.638.369.93Notes:1.16-bit up/down, resettable binary counter (one counter per function block).Recommended Operating ConditionsDC Electrical Characteristics (Over Recommended Operating Conditions)Absolute Maximum RatingsSymbol DescriptionValue Units V CC Supply voltage relative to ground –0.5 to 2.0V V CCIO Supply voltage for output drivers –0.5 to 4.0V V JTAG (2)JTAG input voltage limits –0.5 to 4.0V V CCAUX JTAG input supply voltage –0.5 to 4.0V V IN (1)Input voltage relative to ground –0.5 to 4.0V V TS (1)Voltage applied to 3-state output –0.5 to 4.0V T STG (3)Storage Temperature (ambient)–65 to +150°C T JJunction Temperature+150°CNotes:1.Maximum DC undershoot below GND must be limited to either 0.5V or 10 mA, whichever is easiest to achieve. During transitions,the device pins may undershoot to –2.0v or overshoot to +4.5V, provided this over or undershoot lasts less than 10ns and with the forcing current being limited to 200 mA.2.Valid over commercial temperature range.3.For soldering guidelines and thermal considerations, see the Device Packaging information on the Xilinx website. For Pb freepackages, see XAPP427.Symbol ParameterMin Max Units V CC Supply voltage for internal logic and input buffersCommercial T A = 0°C to +70°C 1.7 1.9V Industrial T A = –40°C to +85°C1.7 1.9V V CCIOSupply voltage for output drivers @ 3.3V operation 3.0 3.6V Supply voltage for output drivers @ 2.5V operation 2.3 2.7V Supply voltage for output drivers @ 1.8V operation 1.7 1.9V Supply voltage for output drivers @ 1.5V operation1.4 1.6V V CCAUXJTAG programming pins1.73.6VSymbol ParameterTest ConditionsTypical Max.Units I CCSB Standby current Commercial V CC = 1.9V, V CCIO = 3.6V 2290μA I CCSB Standby current Industrial V CC = 1.9V, V CCIO = 3.6V38150μA I CC (1)Dynamic current f = 1 MHz -0.25mA f = 50 MHz - 2.5mA C JTAG JTAG input capacitance f = 1 MHz -10pF C CLK Global clock input capacitance f = 1 MHz -12pF C IO I/O capacitance f = 1 MHz-10pF I IL (2)Input leakage current V IN = 0V or V CCIO to 3.9V -+/-1μA I IH (2)I/O High-Z leakageV IN = 0V or V CCIO to 3.9V-+/-1μANotes:1.16-bit up/down resettable binary counter (one per Function Block) tested at V CC = V CCIO = 1.9V.2.See Quality and Reliability section of the CoolRunner-II family data sheet.LVCMOS 3.3V and LVTTL 3.3V DC Voltage SpecificationsSymbol Parameter Test Conditions Min.Max.UnitsV CCIO Input source voltage 3.0 3.6VV IH High level input voltage2 3.9VV IL Low level input voltage–0.30.8VV OH High level output voltage I OH = –8 mA, V CCIO = 3V V CCIO – 0.4V-VI OH = –0.1 mA, V CCIO = 3V V CCIO – 0.2V-VV OL Low level output voltage I OL = 8 mA, V CCIO = 3V-0.4VI OL = 0.1 mA, V CCIO = 3V-0.2VLVCMOS 2.5V DC Voltage SpecificationsSymbol Parameter Test Conditions Min.Max.UnitsV CCIO Input source voltage 2.3 2.7VV IH High level input voltage 1.7V CCIO + 0.3(1)VV IL Low level input voltage–0.30.7VV OH High level output voltage I OH = –8 mA, V CCIO = 2.3V V CCIO – 0.4V-VI OH = –0.1 mA, V CCIO = 2.3V V CCIO – 0.2V-VV OL Low level output voltage I OL = 8 mA, V CCIO = 2.3V-0.4VI OL = 0.1mA, V CCIO = 2.3V-0.2V(1) The V IH Max value represents the JEDEC specification for LVCMOS25. The CoolRunner-II input buffer can tolerate up to 3.9V without physical damage.LVCMOS 1.8V DC Voltage SpecificationsSymbol Parameter Test Conditions Min.Max.Units V CCIO Input source voltage 1.7 1.9VV IH High level input voltage0.65 x V CCIO V CCIO + 0.3(1)VV IL Low level input voltage–0.30.35 x V CCIO VV OH High level output voltage I OH = –8 mA, V CCIO = 1.7V V CCIO – 0.45-VI OH = –0.1 mA, V CCIO = 1.7V V CCIO – 0.2-VV OL Low level output voltage I OL = 8 mA, V CCIO = 1.7V-0.45VI OL = 0.1 mA, V CCIO = 1.7V-0.2V(1) The V IH Max value represents the JEDEC specification for LVCMOS18. The CoolRunner-II input buffer can tolerate up to 3.9V without physical damage.LVCMOS1.5V DC Voltage Specifications(1)Symbol Parameter Test Conditions Min.Max.Units V CCIO Input source voltage 1.4 1.6VV T+Input hysteresis threshold voltage0.5 x V CCIO0.8 x V CCIO VV T-0.2 x V CCIO0.5 x V CCIO VV OH High level output voltage I OH = –8 mA, V CCIO = 1.4V V CCIO – 0.45-VI OH = –0.1 mA, V CCIO = 1.4V V CCIO – 0.2-V4DS310 (v2.0) March 8, 2007Schmitt Trigger Input DC Voltage SpecificationsAC Electrical Characteristics Over Recommended Operating ConditionsV OLLow level output voltageI OL = 8 mA, V CCIO = 1.4V -0.4V I OL = 0.1 mA, V CCIO = 1.4V-0.2VNotes:1.Hysteresis used on 1.5V inputs.Symbol ParameterTest ConditionsMin.Max.Units V CCIO Input source voltage1.4 3.9V V T+Input hysteresis threshold voltage0.5 x V CCIO 0.8 x V CCIO V V T-0.2 x V CCIO0.5 x V CCIOVSymbol Parameter-4-6Units Min.Max.Min.Max.T PD1Propagation delay single p-term - 3.8- 5.5ns T PD2Propagation delay OR array - 4.0- 6.0ns T SUD Direct input register clock setup time 1.7- 2.2-ns T SU1Setup time fast (single p-term) 1.9- 2.6-ns T SU2Setup time (OR array) 2.1- 3.1-ns T HD Direct input register hold time 0.0-0.0-ns T H P-term hold time 0.0-0.0-ns T CO Clock to output - 3.7- 4.7ns F TOGGLE (1)Internal toggle rate-500-300MHz F SYSTEM1(2)Maximum system frequency -323-200MHz F SYSTEM2(2)Maximum system frequency -303-182MHz F EXT1(3)Maximum external frequency -179-137MHz F EXT2(3)Maximum external frequency-172-128MHz T PSUD Direct input register p-term clock setup time 0.4-0.9-ns T PSU1P-term clock setup time (single p-term)0.6- 1.3-ns T PSU2P-term clock setup time (OR array)0.8- 1.8-ns T PHD Direct input register p-term clock hold time 1.5- 1.6-ns T PH P-term clock hold 1.3- 1.2-ns T PCO P-term clock to output- 5.0- 6.0ns T OE /T OD Global OE to output enable/disable - 4.7- 5.5ns T POE /T POD P-term OE to output enable/disable- 6.2- 6.7ns T MOE /T MOD Macrocell driven OE to output enable/disable - 6.2- 6.9ns T PAO P-term set/reset to output valid - 5.5- 6.8ns T AO Global set/reset to output valid - 4.5- 5.5ns T SUEC Register clock enable setup time 2.0- 3.0-ns T HEC Register clock enable hold time 0.0-0.0-ns T CW Global clock pulse width High or Low 1.4- 2.2-ns T PCWP-term pulse width High or Low4.0- 6.0-nsSymbol ParameterTest ConditionsMin.Max.Units6DS310 (v2.0) March 8, 2007T APRPW Asynchronous preset/reset pulse width (High or Low)4.0- 6.0-ns T CONFIG (4)Configuration time-50-50μsNotes:1.F TOGGLE is the maximum clock frequency to which a T-Flip Flop can reliably toggle (see the CoolRunner-II family data sheet).2.F SYSTEM1 (1/T CYCLE ) is the internal operating frequency for a device fully populated with one 16-bit counter through one p-term permacrocell while F SYSTEM2 is through the OR array .3.F EXT1 (1/T SU1+T CO ) is the maximum external frequency using one p-term while F EXT2 is through the OR array .4.Typical configuration current during T CONFIG is 500 μA.Symbol Parameter-4-6Units Min.Max.Min.Max.Internal Timing ParametersSymbol Parameter(1)-4-6Units Min.Max.Min.Max.Buffer DelaysT IN Input buffer delay- 1.3- 1.7ns T DIN Direct register input delay- 1.5- 2.4ns T GCK Global Clock buffer delay- 1.3- 2.0ns T GSR Global set/reset buffer delay- 1.6- 2.0ns T GTS Global 3-state buffer delay- 1.1- 2.1ns T OUT Output buffer delay- 1.8- 2.0ns T EN Output buffer enable/disable delay- 2.9- 3.4ns P-term DelaysT CT Control term delay- 1.3- 1.6ns T LOGI1Single p-term delay adder-0.4- 1.1ns T LOGI2Multiple p-term delay adder-0.2-0.5ns Macrocell DelayT PDI Input to output valid-0.3-0.7ns T LDI Setup before clock (transparent latch)- 1.5- 2.5ns T SUI Setup before clock 1.5- 1.8-ns T HI Hold after clock0.0-0.0-ns T ECSU Enable clock setup time 0.7- 1.7-ns T ECHO Enable clock hold time 0.0-0.0-ns T COI Clock to output valid-0.6-0.7ns T AOI Set/reset to output valid- 1.1- 1.5ns Feedback DelaysT F Feedback delay-0.6- 1.4ns T OEM Macrocell to global OE delay-0.7-0.8ns I/O Standard Time Adder Delays 1.5V CMOST HYS15Hysteresis input adder- 3.0- 4.0ns T OUT15Output adder-0.8- 1.0ns T SLEW15Output slew rate adder- 4.0- 5.0ns I/O Standard Time Adder Delays 1.8V CMOST HYS18Hysteresis input adder- 3.0- 4.0ns T OUT18Output adder-0.0-0.0ns T SLEW Output slew rate adder- 4.0- 5.0ns8DS310 (v2.0) March 8, 2007Switching CharacteristicsAC Test CircuitI/O Standard Time Adder Delays 2.5V CMOS T IN25Standard input adder -0.5-0.6ns T HYS25Hysteresis input adder - 3.0- 4.0ns T OUT25Output adder-0.6-0.7ns T SLEW25Output slew rate adder- 4.0- 5.0ns I/O Standard Time Adder Delays 3.3V CMOS/TTL T IN33Standard input adder -0.5-0.6ns T HYS33Hysteresis input adder - 3.0- 4.0ns T OUT33Output adder- 1.0- 1.2ns T SLEW33Output slew rate adder- 4.0- 5.0nsNotes:1. 1.5 ns input pin signal rise/fall.Internal Timing Parameters (Continued)Symbol Parameter (1)-4-6Units Min.Max.Min.Max.Figure 2: Derating Curve for T PDFigure 3: AC Load CircuitTypical I/O Output CurvesFigure 4: Typical I/V Curve for XC2C32APin DescriptionsFunction Block Macrocell QFG32PC44VQ44CP56I/O Bank 114438F1Bank 2124337E3Bank 2134236E1Bank 2 1(GTS1)434034D1Bank21(GTS0)523933C1Bank21(GTS3)613832A3Bank 21(GTS2)7323731A2Bank 21(GSR)8313630B1Bank 221 9303529A1Bank21 10293428C4Bank111283327C5Bank2112242923C8Bank21132822A10Bank2114232721B10Bank21152620C10Bank21162519E8Bank2215139G1Bank 122240F3Bank123341H1Bank124442G3Bank1 2(GCK0)56543J1Bank 12(GCK1)67644K1Bank 12(GCK2)7871K2Bank 110DS310 (v2.0) March 8, 2007XC2C32A Global, JTAG, Power/Ground and No Connect Pins28982K3Bank1291093H3Bank1210115K5Bank 1211126H5Bank 121213148H8Bank 1213171812K8Bank1214181913H10Bank1215192014G10Bank 12162216F10Bank1Notes:1.GTS = global output enable, GSR = global set reset, GCK = global clock2.GTS, GSR, and GCK pins can also be used for general purpose I/O.Pin TypeQFG32PC44(1)VQ44(1)CP56(1)TCK 161711K10TDI 14159J10TDO 253024A6TMS 151610K9Input Only22 (bank 2)24 (bank 2)18 (bank 2)D10 (bank 2)V CCAUX (JTAG supply voltage)44135D3Power internal (V CC )Power bank 1 I/O (V CCIO1)Power bank 2 I/O (V CCIO2)202115G812137H6273226C6Ground 11, 21, 2610,23,314,17,25H4, F8, C7No connects--K4, K6, K7, H7, E10, A7, A9, D8, A5, A8,A4, C3Total user I/O (includes dual function pins)21333333Notes:1.All packages pin compatible with larger macrocell densitiesPin Descriptions (Continued)Function BlockMacrocellQFG32PC44VQ44CP56I/O BankOrdering InformationPart Number Pin/BallSpacingθJA(C/Watt)θJC(C/Watt)Package TypePackage BodyDimensions I/OComm.(C)Ind. (I)(1)XC2C32A-4QFG32C0.5mm35.524.0Quad Flat No Lead;Pb-free5mm x 5mm21CXC2C32A-6QFG32C0.5mm35.524.0Quad Flat No Lead;Pb-free5mm x 5mm21CXC2C32A-4PC44C 1.27mm55.135.3Plastic Leaded ChipCarrier16.5mm x 16.5mm33CXC2C32A-6PC44C 1.27mm55.135.3Plastic Leaded ChipCarrier16.5mm x 16.5mm33CXC2C32A-4VQ44C0.8mm47.78.2Very Thin Quad FlatPack10mm x 10mm33CXC2C32A-6VQ44C0.8mm47.78.2Very Thin Quad FlatPack10mm x 10mm33C XC2C32A-4CP56C0.5mm66.014.9Chip Scale Package6mm x 6mm33C XC2C32A-6CP56C0.5mm66.014.9Chip Scale Package6mm x 6mm33C XC2C32A-4PCG44C 1.27mm55.135.3Plastic Leaded ChipCarrier; Pb-free16.5mm x 16.5mm33CXC2C32A-6PCG44C 1.27mm55.135.3Plastic Leaded ChipCarrier; Pb-free16.5mm x 16.5mm33CXC2C32A-4VQG44C0.8mm47.78.2Very Thin Quad FlatPack; Pb-free10mm x 10mm33CXC2C32A-6VQG44C0.8mm47.78.2Very Thin Quad FlatPack; Pb-free10mm x 10mm33CXC2C32A-4CPG56C0.5mm66.014.9Chip Scale Package;Pb-free6mm x 6mm33CXC2C32A-6CPG56C0.5mm66.014.9Chip Scale Package;Pb-free6mm x 6mm33CXC2C32A-6QFG32I0.5mm35.524.0Quad Flat No Lead;Pb-free5mm x 5mm21IXC2C32A-6PC44I 1.27mm55.135.3Plastic Leaded ChipCarrier16.5mm x 16.5mm33IXC2C32A-6VQ44I0.8mm47.78.2Very Thin Quad FlatPack10mm x 10mm33I XC2C32A-6CP56I0.5mm66.014.9Chip Scale Package6mm x 6mm33I XC2C32A-6PCG44I 1.27mm55.135.3Plastic Leaded ChipCarrier; Pb-free16.5mm x 16.5mm33I12DS310 (v2.0) March 8, 2007Device Part MarkingFigure 5: Sample Package with Part MarkingNote: Due to the small size of chip scale and quad flat no lead packages, the complete ordering part number cannot be included on the package marking. Part marking on chip scale and quad flat no lead packages by line are:•Line 1 = X (Xilinx logo) then truncated part number •Line 2 = Not related to device part number •Line 3 = Not related to device part number•Line 4 = Package code, speed, operating temperature, three digits not related to device part number. Package codes: C3 = CP56, C4 = CPG56, Q1 = QFG32.XC2C32A-6VQG44I 0.8mm 47.78.2Very Thin Quad Flat Pack; Pb-free 10mm x 10mm 33I XC2C32A-6CPG56I0.5mm66.014.9Chip Scale Package;Pb-free6mm x 6mm33INotes:1. C = Commercial (T= 0°C to +70°C); I = Industrial (T = –40°C to +85°C)Part Number Pin/Ball Spacing θJA (C/Watt)θJC (C/Watt)Package Type Package Body Dimensions I/O Comm. (C)Ind. (I)(1)Figure 6: QFG32 PackageFigure 7: VQ44 PackageFigure 8: PC44 PackageFigure 9: CP56 PackageWarranty DisclaimerTHESE PRODUCTS ARE SUBJECT TO THE TERMS OF THE XILINX LIMITED WARRANTY WHICH CAN BE VIEWED AT /warranty.htm. THIS LIMITED WARRANTY DOES NOT EXTEND TO ANY USE OF THE PRODUCTS IN AN APPLICATION OR ENVIRONMENT THAT IS NOT WITHIN THE SPECIFICATIONS STATED ON THE THEN-CURRENT XILINX DATA SHEET FOR THE PRODUCTS. PRODUCTS ARE NOT DESIGNED TO BE FAIL-SAFE AND ARE NOT WARRANTED FOR USE IN APPLICATIONS THAT POSE A RISK OF PHYSICAL HARM OR LOSS OF LIFE. USE OF PRODUCTS IN SUCH APPLICATIONS IS FULLY AT THE RISK OF CUSTOMER SUBJECT TO APPLICABLE LAWS AND REGULATIONS.14DS310 (v2.0) March 8, 2007Additional InformationAdditional information is available for the following CoolRunner-II topics:•XAPP784: Bulletproof CPLD Design Practices •XAPP375: Timing Model•XAPP376: Logic Engine•XAPP378: Advanced Features•XAPP382: I/O Characteristics•XAPP389: Powering CoolRunner-II•XAPP399: Assigning VREF Pins To access these and all application notes with their associ-ated reference designs, click the following link and scroll down the page until you find the document you want: CoolRunner-II Data Sheets and Application Notes Device PackagesRevision HistoryThe following table shows the revision history for this document.Date Version Revision6/15/04 1.0Initial Xilinx release.8/30/04 1.1Pb-free documentation10/01/04 1.2Add Asynchronous Preset/Reset Pulse Width specification to AC Electrical Characteristics.11/08/04 1.3Product Release. No changes to documentation.11/22/04 1.4Changes to output enable/disable specifications; changes to I CCSB.02/17/05 1.5Changes to f TOGGLE, t SLEW25, and t SLEW3303/07/05 1.6Improvement of pin-to-pin logic delay, page 1. Modifications to Table 1, IOSTANDARDs.06/28/05 1.7Move to Product Specification. Change to T IN25, T OUT25, T IN33, and T OUT33.03/20/06 1.8Add Warranty Disclaimer. Add note to Pin Descriptions that GCK, GSR, and GTS pins can alsobe used for general purpose I/O.02/15/07 1.9Change to V IH specification for 2.5V and 1.8V LVCMOS. Change to T OEM for -4 speedgrade.03/08/07 2.0Fixed typo in note for V IL for LVCMOS18; removed note for V IL for LVCMOS33.16DS310 (v2.0) March 8, 2007。

ptuning原理

ptuning原理Ptuning原理解析简介Ptuning是一种被广泛应用于创造性工作中的原理，它能够帮助创作者更好地展现他们的创造力和激发灵感。

本文将从浅入深地解释Ptuning原理的相关概念和使用方法。

什么是Ptuning原理Ptuning原理是一种通过探索、挖掘和利用创造性的潜力来帮助创作者提高创作效果的方法。

Ptuning的全称是Potential tuning，意为潜能调整。

它的基本理念是，每个人都有无限的创造潜力，只需要通过适当的方法和技巧来激发和调整这些潜能，就能取得更好的创造成果。

Ptuning原理的关键要素Ptuning原理包含以下几个关键要素：1. 探索潜能Ptuning的第一步是探索个体的潜能。

这意味着创作者需要从自身出发，认识自己的优点、长处以及可能未被充分发掘的潜力。

这种探索通常涉及反思、自我评估和尝试新的创作方法等步骤。

2. 确定目标Ptuning的下一步是明确创作者的目标。

无论是写作、绘画还是设计，创作者都应该明确自己希望实现的创作目标。

这些目标可以是短期的，也可以是长期的，它们将作为Ptuning的指导方针。

3. 调整策略Ptuning的关键环节是根据创作者的目标调整创作策略。

这意味着创作者需要不断地尝试不同的方法、技巧和创造性工具，以找到最适合其目标的创作方式。

调整策略可能需要时间和经验，但它将对创作的效果产生重要影响。

4. 实践与反馈Ptuning的最后一步是通过实践和反馈不断优化创作过程和成果。

创作者应该不断地进行实验和尝试，并从中获得反馈，以便进一步优化创作。

这种循环将帮助创作者不断成长和提升创作水平。

如何应用Ptuning原理应用Ptuning原理可以帮助创作者更好地实现他们的创造潜力。

以下是一些可以采用的方法：•思维导图：使用思维导图可以帮助创作者整理和展示创作思路，激发创造力的火花。

•尝试不同的媒介：创作者可以尝试跨界创作，将不同的媒介融合在一起，例如在绘画作品中加入文字元素。

sharding-jdbc 雪花片算法 java 调用

sharding-jdbc 雪花片算法java 调用1. 引言1.1 概述本文将介绍sharding-jdbc与雪花片算法的结合应用。

sharding-jdbc是一个轻量级的Java框架，用于实现关系型数据库的分库分表功能，旨在解决数据量过大时数据库性能下降的问题。

而雪花片算法是一种用于生成分布式系统中唯一ID的算法，其生成的ID具有趋势递增、可排序、长度较短等特点。

本文将详细介绍雪花片算法的原理和实现细节，并讨论其在实际应用中的优势和适用场景。

1.2 文章结构本文共分为五个部分。

首先是引言部分，概述了文章内容和结构；接着是雪花片算法简介，阐述了该算法的原理、实现细节以及应用场景；然后介绍了sharding-jdbc这个框架以及其核心组件功能；紧接着是主要内容——java调用sharding-jdbc实现雪花片算法数据分片操作；最后通过结论与展望总结全文，并对未来发展进行展望。

1.3 目的本文旨在帮助读者了解并学习如何使用sharding-jdbc框架与雪花片算法相结合，实现数据分片操作。

通过本文的阅读，读者可以了解到雪花片算法的原理和优点，并学习如何使用sharding-jdbc框架来实现雪花片算法的数据分片操作。

同时，本文还对未来发展进行了展望，希望能够为读者提供一些关于该领域未来趋势的思考和启示。

2. 雪花片算法简介2.1 算法原理雪花片算法（Snowflake Algorithm）是一种用于生成分布式唯一ID的算法。

它的原理基于时间戳、工作机器ID和序列号的组合。

具体地说，Snowflake算法通过将时间戳设置为高位，将机器ID设置为中间位，将序列号设置为低位来生成一个64位的唯一ID。

在Snowflake算法中，时间戳表示了该ID的生成时间。

由于采用时间戳作为高位，因此Snowflake可以保证在相同的机器和相同的毫秒内生成的ID是递增有序的。

机器ID是用以区分不同节点/机器之间生成的ID。

PD30CNB20传感器系列数据表说明书

Specifications are subject to change without notice (13.06.2016)1Product Description The PD30CNB20 sensor family comes in a compact 10 x 30 x 20 mm reinforced PMMA/ABS housing. The sensors are useful in applications where high-ac-curacy detection as well as small size is required. Compact housing and high power LED for excellent performance-size ratio.The potentiometer function for adjustment of the sen-sitivity makes the sensors highly flexible. The output type is preset (NPN or PNP), and the output switching function is NO and NC out-put.• Miniature sensor range • Range: 200 mm• Sensitivity adjustment by potentiometer • Modulated, red light 625 nm • Supply voltage: 10 to 30 VDC• Output: 100 mA, NPN or PNP preset • Make and break switching function• LED indication for output, stability and power ON• Protection: reverse polarity, short circuit and transients • Cable and plug versions • Excellent EMC performance • Exelent colour matchingPhotoelectricsDiffuse-reflective, Background Suppression Type PD30CNB20....SAType SelectionHousing Range Connection Ordering no.Ordering no. W x H x D S n NPNPNPMake and break switching Make and break switching10 x 30 x 20 mm 200 mm Cable PD 30 CNB 20 NASA PD 30 CNB 20 PASA 10 x 30 x 20 mm200 mmPlugPD 30 CNB 20 NAM5SAPD 30 CNB 20 PAM5SASpecificationsEN 60947-5-22Specifications are subject to change without notice (13.06.2016)PD30CNB20....SASpecifications (cont.)Operation DiagramPower supply ObjectOFF ONPresent Break Output (N.C.)Make Output (N.O.)OFF ONOFFONTv = Power ON delaySensing ConditionsDetection Diagram -4-3-2-101234(m m )50100150200250Sensing range (mm)-0,16 -0,12-0,08 -0,04 0,040,08 0,12 0,16(I n c h e s )2,03,95,97,99,8Sensing range (inches)0,00Specifications are subject to change without notice (13.06.2016)3Delivery Contents• Photoelectric switch: PD30CNB20 ...• Screwdriver• Packaging: Plastic bagPD30CNB20....SADimensionsInstallation HintsWiring DiagramsAccessories• Mounting bracket APD30-MB1 or APD30-MB2 to be purchased separately.。

基于SE_注意力CycleGAN_的蓝印花布单纹样自动生成

理论与方法丝绸ＪＯＵＲＮＡＬＯＦＳＩＬＫ基于ＳＥ注意力ＣｙｃｌｅＧＡＮ的蓝印花布单纹样自动生成ＳｉｎｇｌｅｐａｔｔｅｒｎａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮ冉二飞１ꎬ２ꎬ贾小军１ꎬ２ꎬ喻擎苍１ꎬ谢㊀昊１ꎬ２ꎬ陈卫彪２ꎬ３(１.浙江理工大学计算机科学与技术学院(人工智能学院)ꎬ杭州３１００１８ꎻ２.嘉兴学院信息科学与工程学院ꎬ浙江嘉兴３１４００１ꎻ３.浙江师范大学数学与计算机科学学院ꎬ浙江金华３２１００４)摘要:根据蓝印花布纹样的风格特征ꎬ文章提出一种端到端的蓝印花布纹样自动生成方法ꎬ实现简笔画图像向蓝印花布单纹样的自动迁移ꎮ针对蓝印花布的抽象风格和小数据集问题ꎬ重新构造ＣｙｃｌｅＧＡＮ生成网络中的编码器和解码器ꎬ使用ＳＥ(ｓｑｕｅｅｚｅａｎｄｅｘｃｉｔａｔｉｏｎ)注意力模块和残差模块与原始的卷积模块串联ꎬ提高特征提取能力和网络学习能力ꎮ同时减少生成网络中转换器的残差块层数ꎬ降低过拟合ꎮ实验结果表明ꎬ基于ＳＥ注意力ＣｙｃｌｅＧＡＮ网络方法自动生成的蓝印花布新纹样主观性上更贴合原始风格ꎬ与原图更加接近ꎬ有助于蓝印花布的数字化传承和创新ꎮ关键词:蓝印花布ꎻＳＥ注意力ꎻ风格迁移ꎻＣｙｃｌｅＧＡＮꎻ单纹样ꎻ半监督学习ꎻ图像生成中图分类号:ＴＳ９４１.２ꎻＴＰ３９１.７㊀㊀㊀㊀文献标志码:Ａ㊀㊀㊀㊀文章编号:１００１７００３(２０２４)０１００３１０７ＤＯＩ:１０.３９６９/ｊ.ｉｓｓｎ.１００１￣７００３.２０２４.０１.００４收稿日期:２０２３０５２５ꎻ修回日期:２０２３１２０１作者简介:冉二飞(１９９８)ꎬ男ꎬ硕士研究生ꎬ研究方向为计算机视觉与图像处理ꎮ通信作者:贾小军ꎬ教授ꎬｘｊｊｉａｄ＠ｓｉｎａ.ｃｏｍꎮ㊀㊀蓝印花布源于唐宋ꎬ盛于明清ꎬ是中国传统的民间手工艺品ꎮ作为首批列入国家级非物质文化遗产名录的民间传统工艺ꎬ其使用简单淳朴的蓝白两色ꎬ创造出绚丽多姿的艺术世界ꎬ因其纹样设计风格特征鲜明㊁线断意连而闻名于世ꎮ目前对蓝印花布的研究ꎬ大都是从纹样的视觉语义寓意ꎬ历史发展和制作工艺角度进行[１￣３]ꎬ而对于纹样的自动生成技术研究略显单薄ꎬ且往往不够简单实用[４￣５]ꎮ随着深度学习的发展ꎬ尤其是图像风格迁移研究的进展ꎬ提供了扩展蓝印花布纹样的新思路ꎬ即将其他风格的简单图像转化为蓝印花布风格的纹样ꎬ从而大大提高蓝印花布设计的效率ꎬ有利于创新蓝印花布纹样ꎮ图像的风格迁移是两个不同域中图像的转换ꎬ具体来说就是提供一种风格图像ꎬ将任意一张图像转化为这种风格ꎬ并尽可能保留原图像的内容ꎮ如今实现风格迁移的方法多种多样ꎬ大致分为基于神经网络的图像风格迁移和基于对抗生成网络的图像风格迁移ꎮＧａｔｙｓ等[６]首先将深度学习运用在风格迁移任务上ꎬ利用Ｇｒａｍ矩阵将图像表示为内容和风格两部分ꎬ通过图像重建使内容图的Ｇｒａｍ矩阵逼近风格图的Ｇｒａｍ矩阵ꎮ这种方法可以生成风格图像ꎬ但是收敛速度慢ꎬ渲染时间长ꎮＷａｎｇ等[７]为了减少伪影ꎬ引入了相似性损失函数ꎬ添加了一个后处理细化步骤ꎮ其方法可以稳定地对摄影作品的图像进行风格转换ꎮＬｕａｎ等[８]约束卷积神经网络从输入到输出的变换ꎬ使其表示为颜色空间中的局部仿射ꎬ实现了如天气㊁季节等多种场景的艺术风格转换ꎬ但是风格迁移的效率低下ꎬ效果一般ꎮＪｏｈｎｓｏｎ等[９]使用在ＩｍａｇｅＮｅｔ上预训练的ＶＧＧＮｅｔ简化损失函数计算过程ꎬ使效率得到较大提升ꎬ但该方法需要构造复杂的损失函数ꎮＧｏｏｄｆｅｌｌｏｗ等[１０]提出了生成对抗网络理论(ＧＡＮ)ꎬＧＡＮ为图像风格转换提供了新的思路ꎬ掀起了新的研究热潮ꎮＭｉｒｚａ等[１１]提出了带有条件约束的ｃＧＡＮ(ｃｏｎｄｉｔｉｏｎａｌＧＡＮ)ꎬ该模型通过对输入图像额外增加一个条件标签ꎬ来引导模型生成方向ꎮＩｓｏｌａ等[１２]提出了Ｐｉｘ２Ｐｉｘ算法模型ꎬ通过图像作为输入来进行图像风格转换ꎬ而不是传统的噪音ꎬ大大提升了生成图像的可控性ꎮｃＧＡＮ和Ｐｉｘ２Ｐｉｘ都需要配对的数据集ꎬ但是在很多情况下ꎬ并没有完美的成对数据集ꎮＺｈｕ等[１３]提出了ＣｙｃｌｅＧＡＮ算法模型ꎬ摆脱了配对训练数据集的限制要求ꎬ使用半监督的方式实现不同风格之间的图像转换ꎬ如将马转化为斑马㊁春夏秋冬的转换㊁油画和真实图像的转换等ꎬ但ＣｙｃｌｅＧＡＮ在几何形状改变方面表现不佳ꎮＣｈｅｎ等[１４]提出ＣａｒｔｏｏｎＧＡＮꎬ用于将现实图像转化为漫画风格ꎮ该模型在ＣｙｃｌｅＧＡＮ的基础上针对卡通图像加入边缘对抗损失ꎬ使生成的图像具有漫画图像一样的清晰边缘ꎬ但在处理现实人脸到漫画人脸这类形变较大的转换时效果不佳ꎮ以上模型大多只对于图像纹理与色调的风格进行转换ꎬ忽略了几何形变方面的风格转换ꎮ蓝印花布的风格有很强的抽象特性ꎬ为了稳定生成蓝印花布的纹样ꎬ针对上述图像风格化存在问题ꎬ本文使用ＳＥ注意力机制改进ＣｙｃｌｅＧＡＮ模型来１３Ｖｏｌ.６１㊀Ｎｏ.１ＳｉｎｇｌｅｐａｔｔｅｒｎａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮ实现蓝印花布纹样图像的风格迁移ꎮ对生成网络中的编码器和解码器进行重构ꎬ提升ＣｙｃｌｅＧＡＮ模型几何形变方面的能力ꎬ使得生成结果贴近蓝印花布纹样的抽象风格ꎮ１㊀ＣｙｃｌｅＧＡＮ原理ＧＡＮ模型由生成网络和判别网络组成ꎬ其目标是让生成网络的模型学会一种映射ꎬ使得原始域的数据分布拟合目标域的数据分布ꎮ在训练过程中ꎬ生成网络需要生成伪造的样本使判别网络判断为真ꎬ而判别网络则要尽力判别输入样本是真实样本还是生成的伪造样本ꎬ两个神经网络在这个对抗中不断优化ꎬ最后网络模型能够输出接近样本分布的数据ꎮＧＡＮ的优化目标可以看成是一个极大极小博弈ꎬ在训练最后的生成网络和判别网络之间实现纳什均衡ꎬ使得生成网络能够生成接近样本数据分布特征的目标数据ꎮ其目标方程[１０]为:ｍｉｎＧ㊀ｍａｘＤＶ(ＤꎬＧ)＝Ｅｘ~ｐｄａｔａ[ｌｏｇ(Ｄ(ｘ))]＋Ｅｚ~ｐｚ[ｌｏｇ(１－Ｄ(Ｇ(ｚ)))](１)式中:Ｇ表示生成网络ꎬＤ表示判别网络ꎬＰｄａｔａ表示真实样本ｘ的分布ꎬＰｚ表示输入的噪声ｚ的分布ꎬＶ表示损失函数ꎬＥ表示数学期望ꎬＧ(ｚ)表示生成网络Ｇ根据噪声ｚ生成的假图像ꎬＤ(ｘ)和Ｄ(Ｇ(ｚ))分别表示判别器Ｄ判断真实样本ｘ和假图像Ｇ(ｚ)是真实样本分布的概率ꎮＧＡＮ模型在跨域图像风格转换任务上存在一个缺陷ꎬ生成网络可能会把原始域映射到目标域上的子集ꎬ甚至有可能将原始域全部映射到一张图像上ꎬ而判别网络只关注生成图像是否属于目标域ꎬ所以仅通过单独的对抗损失ꎬ无法达到将原始域映射到目标域的结果ꎮ对于这个问题ꎬＣｙｃｌｅＧＡＮ没有采用像Ｐｉｘ２Ｐｉｘ算法使用严格配对数据集的做法ꎬ而是使用循环一致性损失解决这一问题ꎮＣｙｃｌｅＧＡＮ模型通过第一个生成网络Ｇ(ｘ２ｙ)ꎬ将输入的Ｘ域图像转换成Ｙ域ꎬ然后通过第二个生成网络Ｆ(ｙ２ｘ)转换回来ꎬ将原始域中的数据经过两次转换后ꎬ转换回来的图像应与原始输入尽量相同ꎮ同样地ꎬ对于Ｙ域的图像通过Ｆ(ｙ２ｘ)和Ｇ(ｘ２ｙ)重新生成伪造的Ｙ域图像与原始输入的Ｙ域图像进行比较ꎮ通过这种方式解决了Ｘ域可能都映射到Ｙ域同一张图像的情况ꎮＣｙｃｌｅＧＡＮ模型有两个生成网络和两个判别网络ꎬ其中两个生成对抗网络中的生成网络共享权重ꎮＣｙｃｌｅＧＡＮ模型在结构上像一个环形网络ꎬ从Ｘ域向Ｙ域转换的ＧＡＮ网络结构如图１所示ꎮＣｙｃｌｅＧＡＮ循环一致性损失函数计算方法[１３]为:Ｌｃｙｃ(ＧꎬＦ)＝Ｅｘ~ｐｄａｔａ(ｘ)Ｆ(Ｇ(ｘ)－ｘ)１＋Ｅｙ~ｐｄａｔａ(ｙ)Ｆ(Ｇ(ｙ)－ｙ)１(２)对抗损失由两部分组成ꎬ分别对应两个单独的ＧＡＮ模型:ＬＧＡＮ(ＧꎬＤｙꎬＸꎬＹ)＝Ｅｙ~ｐｄａｔａ(ｙ)[ｌｏｇＤｙ(ｙ)]＋Ｅｘ~ｐｄａｔａ(ｘ)[ｌｏｇ(１－Ｄｙ(Ｇ(ｘ)))](３)图１㊀ＣｙｃｌｅＧＡＮ网络结构Ｆｉｇ.１㊀ＣｙｃｌｅＧＡＮｎｅｔｗｏｒｋｓｔｒｕｃｔｕｒｅ㊀㊀ＬＧＡＮ(ＦꎬＤｘꎬＹꎬＸ)＝Ｅｘ~ｐｄａｔａ(ｘ)[ｌｏｇＤｘ(ｘ)]＋Ｅｙ~ｐｄａｔａ(ｙ)[ｌｏｇ(１－Ｄｘ(Ｆ(ｙ)))](４)整个环形ＧＡＮ网络的总损失[１３]计算方法为:Ｌ(ＧꎬＦꎬＤｘꎬＤｙ)＝ＬＧＡＮ(ＧꎬＤｙꎬＸꎬＹ)＋ＬＧＡＮ(ＦꎬＤｘꎬＹꎬＸ)＋λＬｃｙｃ(ＧꎬＦ)(５)式中:Ｘ㊁Ｙ分别代表两个数据域ꎬｘ㊁ｙ为两个数据域中的样本数据ꎬＧ为从Ｘ到Ｙ的映射函数ꎬＦ为从Ｙ到Ｘ的映射函数ꎬＤｘ㊁Ｄｙ为判别网络ꎬλ为控制循环一致损失函数的权重ꎮ２㊀改进的ＣｙｃｌｅＧＡＮ网络结构ＣｙｃｌｅＧＡＮ网络实现了无配对图像集之间的风格迁移ꎬ但泛化能力较弱ꎬ当训练图像与测试图像之间差距较大时ꎬ迁移效果不佳ꎮ直接使用ＣｙｃｌｅＧＡＮ进行简笔画向蓝印花布风格纹样转化生成的结果不理想ꎮ针对这一问题对原始ＣｙｃｌｅＧＡＮ的生成网络结构进行改进ꎬＣｙｃｌｅＧＡＮ的生成网络结构由编码器㊁残差结构和解码器组成ꎮ编码器与解码器都是由３个卷积模块组成ꎬ每个模块包含一个卷积层㊁一个实例正则化层及一个Ｒｅｌｕ激活函数ꎮ为了增强ＣｙｃｌｅＧＡＮ网络提取特征的能力ꎬ本文提出了使用残差块结构加上原始卷积模块和ＳＥ注意力模块取代原卷积模块的方法ꎮ２.１㊀注意力机制注意力机制(ａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍ)是解决信息超载问题的一种资源分配方案ꎬ当计算资源有限时ꎬ可以把计算资源分配给更重要的任务ꎮ在神经网络学习过程中ꎬ参数越多模型所存储的信息量就越大ꎬ模型的表达能力也越强ꎬ但这会带来信息过载的问题ꎮ引入注意力机制ꎬ可以在众多的输入信息中聚焦于对当前任务更为关键的信息ꎬ降低对其他信息的关注度ꎬ提高任务处理的效率和准确性ꎮＳＥ(ｓｑｕｅｅｚｅａｎｄｅｘｃｉｔａｔｉｏｎ)模块[１５]在通道维度增加注意力机制ꎬ通过一系列变换操作得到一个权重矩阵ꎬ对原特征进行重构来得到更重要的特征信息ꎬ关键步骤为压缩(ｓｑｕｅｅｚｅ)和激发(ｅｘｃｉｔａｔｉｏｎ)ꎮ通过自动学习的方式ꎬ获取特征通道的重要程度ꎬ以此为每个特征通道赋予不同的权重值ꎬ从而提升对当前任务有用的特征图的通道利用率ꎬ并抑制对当前任务影响不大的通道ꎮＳＥ模块的结构[１５]如图２所示ꎮ２３第６１卷㊀第１期基于ＳＥ注意力ＣｙｃｌｅＧＡＮ的蓝印花布单纹样自动生成图２㊀特征压缩与激发模块结构Ｆｉｇ.２㊀Ｆｅａｔｕｒｅｃｏｍｐｒｅｓｓｉｏｎａｎｄｅｘｃｉｔａｔｉｏｎｍｏｄｕｌｅｓｔｒｕｃｔｕｒｅ㊀㊀实现通道注意力的过程分为压缩(ｓｑｕｅｅｚｅ)ꎬ激发(ｅｘｃｉｔａｔｉｏｎ)和调节(ｓｃａｌｅ)三个步骤ꎮ压缩阶段ꎬ使用全局池化(ｇｌｏｂａｌｐｏｏｌｉｎｇ)操作ꎬ将ＨˑＷˑＣ的特征图沿着通道方向转化成为１ˑ１ˑＣ的特征ꎬ即图２中的Ｆｓｑ过程将每一个二维的特征图变成了一个具有整体卷积感受野的实数Ｚｃꎬ其计算方法为:Ｚｃ＝Ｆｓｑ(ｕｃ)＝１ＨˑＷðＨｉ＝１ðＷｊ＝１ｕｃ(ｉꎬｊ)(６)式中:Ｈ㊁Ｗ为输入特征图的高和宽ꎬｕｃ(ｉꎬｊ)为在ｉ㊁ｊ位置上的特征值ꎮ激发过程的操作Ｆｅｘ( ꎬＷ)是通过两个全连接层来捕捉通道的内部依赖程度参数得到特征通道的裁剪系数ꎬ其计算方法为:ｓ＝Ｆｅｘ(ｚꎬＷ)＝σ(Ｗ２δ(Ｗ１ｚ))(７)式中:Ｗ１和Ｗ２为两个全连接层学习参数ꎬδ为Ｒｅｌｕ函数ꎬσ为Ｓｉｇｍｏｉｄ函数ꎬｚ为上一步求得的１ˑ１ˑＣ的特征向量ꎮ最后是调节权重ꎬ将ｓ视为每个通道的重要程度ꎬ通过Ｆｓｃａｌｅ(.ꎬ.)逐通道加权到先前的特征图上ꎬ完成通道维度上的注意力机制计算ꎮＦｓｃａｌｅ(ｓｃꎬｕｃ)＝ｓｃｕｃ(８)式中:ｓｃ为第Ｃ个通道的裁剪系数ꎬｕｃ为第Ｃ个通道的特征图ꎮ２.２㊀网络结构ＣｙｌｃｅＧＡＮ算法模型是一个循环迭代的网络ꎬ包括两个生成网络Ｇ(ｘ２ｙ)㊁Ｆ(ｙ２ｘ)及两个判别网络Ｄｘ㊁Ｄｙꎮ在本文研究中ꎬＸ代表简笔画数据域ꎬＹ代表蓝印花布纹样数据域ꎮ两个生成网络分别生成简笔画图像和蓝印花布图像ꎮ生成网络通过学习两个图像域之间的映射函数ꎬ最终生成网络Ｇ(ｘ２ｙ)能够使一个简笔画图像转化后获得蓝印花布纹样的风格特点ꎮ２.２.１㊀生成网络生成网络主要由编码器㊁转换器和解码器组成ꎮ本文采用残差块结构㊁原始卷积模块和ＳＥ注意力模块对原ＣｙｃｌｅＧＡＮ模型中编码器和解码器进行改进ꎬ并削减转化器中残差块的数量ꎮ对于新卷积模块(Ｓｅ￣ＣｏｎｖＬａｙｅｒ)ꎬ使用残差模块(ＲｅｓｉｄｕａｌＢｌｏｃｋ)及原始卷积模块(Ｃｏｎｖ３ˑ３ｓｔｒｉｄｅ＝２ｐａｄｄｉｎｇ＝１ꎻＩｎｓｔａｎｃｅＮｏｒｍꎻＲｅｌｕ)加ＳＥ注意力模块(ＳｅＢｌｏｃｋ)取代原始卷积模块ꎬ如图３(ａ)所示ꎮ对于新反卷积模块(ＳｅＤｅＣｏｎｖＬａｙｅｒ)ꎬ使用ＳＥ注意力模块(ＳｅＢｌｏｃｋ)及原始反卷积模块(Ｕｐｓａｍｐｌｅｓｃａｌｅ＿ｆａｃｔｏｒ＝２ꎻＣｏｎｖ３ˑ３ｓｔｒｉｄｅ＝１ｐａｄｄｉｎｇ＝１ꎻＩｎｓｔａｎｃｅＮｏｒｍꎻＲｅｌｕ)加残差模块(ＲｅｓｉｄｕａｌＢｌｏｃｋ)取代原始反卷积模块ꎬ如图３(ｂ)所示ꎮ图３㊀改进的新模块Ｆｉｇ.３㊀Ｎｅｗｉｍｐｒｏｖｅｄｍｏｄｕｌｅｓ改进的整个生成网络结构如图４所示ꎮ首先将输入的２５６ˑ２５６灰度图像送入一个ｐａｄｄｉｎｇ为１ꎬ卷积核大小为３ˑ３的卷积ＩｎｉＣｏｎｖＬａｙｅｒ中进行初始化ꎬ将通道数扩展为１６ꎮ而后将输出送入到ＳｅＣｏｎｖＬａｙｅｒ１中进行特征提取ꎬ并在ＳｅＢｌｏｃｋ中根据式(６~８)进行特征图权重的分配ꎬ得到３２个１２８ˑ１２８的特征向量ꎮ对于得到的特征向量ꎬＳｅＣｏｎｖＬａｙｅｒ２再次进行特征提取和特征图权重的分配ꎬ最终编码器提取出６４个尺寸为６４ˑ６４的特征向量ꎮ由４个残差模块组成的转换器组合提取不同的特征ꎬ将输入转化为目标域的特征向量ꎮ每个残差块中包含２个卷积层㊁２个实例正则化层和１个Ｒｅｌｕ激活层ꎮ解码器为编码器的逆过程ꎬ采用与新卷积模块类似的２个新反卷积层ＳｅＤｅＣｏｎｖＬａｙｅｒ１和ＳｅＤｅＣｏｎｖＬａｙｅｒ２还原图像的低级特征ꎬ得到一个１６ˑ２５６ˑ２５６的特征向量ꎬ最后通过一个１ˑ１卷积层ＦｉｎＣｏｎｖＬａｙｅｒ还原成目标域的图像ꎬ之后将还原的图像输入判别网络判断ꎬ根据式(５)计算梯度ꎬ更新生成网络参数ꎮ３３Ｖｏｌ.６１㊀Ｎｏ.１ＳｉｎｇｌｅｐａｔｔｅｒｎａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮ图４㊀改进的ＣｙｃｌｅＧＡＮ的生成网络结构Ｆｉｇ.４㊀ＩｍｐｒｏｖｅｄＣｙｃｌｅＧＡＮｇｅｎｅｒａｔｉｖｅｎｅｔｗｏｒｋｓｔｒｕｃｔｕｒｅ２.２.２㊀判别网络判别网络的作用为判别生成网络产生的伪样本是否属于目标数据域ꎬ促使生成网络生成更加难以分辨真伪的样本ꎬ其网络结构如图５所示ꎮ本实验的判别网络使用ＰａｔｃｈＧａｎ[１０]判别网络ꎬ由５个卷积模块组成ꎬ前四个模块用于抽取输入图像的特征ꎬ最后一个卷积模块得到一张特征图ꎬ此时特征图上每一个点都有着７０ˑ７０的感受野ꎬ代表了判别网络对于其代表的７０ˑ７０图像块的判别结果ꎮ最后将所有图像块进行判别之后得到的平均结果作为这张输入图像的判别结果ꎬ随后根据式(５)计算梯度ꎬ更新判别网络参数ꎮ图５㊀ＣｙｃｌｅＧＡＮ判别网络结构Ｆｉｇ.５㊀ＣｙｃｌｅＧＡＮｄｉｓｃｒｉｍｉｎａｔｏｒｎｅｔｗｏｒｋｓｔｒｕｃｔｕｒｅ３㊀实验和结果分析３.１㊀环境与数据集实验平台处理器为Ｉｎｔｅｌ(Ｒ)Ｘｅｏｎ(Ｒ)Ｐｌａｔｉｎｕｍ８３５０ＣＣＰＵ＠２.６０ＧＨｚꎬ显卡型号为ＮＡＶＩＤＲＴＸＡ５０００ꎬ２４ＧＢ显存ꎬ操作系统为Ｕｂｕｎｔｕ１８.０４ꎬ使用ＰｙＴｏｒｃｈ深度学习框架ꎬ版本为１.８.１ꎬＣｕｄａ１１.１ꎮ基于改进ＣｙｃｌｅＧＡＮ的风格迁移算法分别选用４１２张动物简笔画和２１６张蓝印花布单纹样图作为实验样本数据集ꎮ训练过程中不同数据集间没有指定配对关系ꎬ其中简笔画数据集通过互联网收集ꎬ蓝印花布单纹样从已有的蓝印花布图像中截取ꎬ两个数据集图像大小统一调整为２５６ˑ２５６像素ꎮ部分数据集图像如图６所示ꎮ３.２㊀训练过程设置循环一致性损失超参数λ值为１０ꎬ批量训练样本数量(ｂａｔｃｈｓｉｚｅ)为１ꎬ训练总轮次(ｅｐｏｃｈ)为４００ꎬ采用ａｄａｍ优化器ꎬ初始学习率(ｌｒ)为０.０００２ꎬ在１００个ｅｐｏｃｈ后使用指数图６㊀训练时的部分数据集Ｆｉｇ.６㊀Ｐａｒｔｉａｌｄａｔａｓｅｔｄｕｒｉｎｇｔｒａｉｎｉｎｇ衰减策略对学习率进行动态调整ꎮ训练过程中的部分实验数据如图７所示ꎮ模型训练５０个ｅｐｏｃｈ时转换效果如图７第二行所示ꎬ此时图像尚没有形成明显的蓝印花布风格ꎬ原始形状保留较为完整ꎬ但此时生成的图像颜色已经变为黑底白纹ꎬ并且组成图像内容的线条已经转化为大小不同的白色块ꎬ有了向蓝印花布风格转化的趋势ꎮ模型训练１００个ｅｐｏｃｈ时转换效果如图７第三行所示ꎬ此时产生的图像仍显得零乱ꎬ产生的色块也较多ꎬ不符合蓝印花布抽象精炼的特点ꎬ但是相比５０个ｅｐｏｃｈ时产生的色块少了许多ꎬ初步形成了蓝印花布的风格ꎮ模型训练４００个ｅｐｏｃｈ时转换效果如图７第四行所示ꎬ此时生成的图像抓住了原始图像中的关键信息ꎬ并将其融合进了蓝印花布的风格里ꎬ图案边界更加清晰ꎬ主体模糊程度明显减弱ꎬ整体风格更为凝实㊁简练ꎬ并在不妨碍整体布局的情况下ꎬ对于过长的线条或弯曲弧度较大的造型产生了合适的断线ꎬ比较符合蓝印花布线断意连的特点ꎮ４３第６１卷㊀第１期基于ＳＥ注意力ＣｙｃｌｅＧＡＮ的蓝印花布单纹样自动生成图７㊀训练过程的转换效果Ｆｉｇ.７㊀Ｃｏｎｖｅｒｓｉｏｎｅｆｆｅｃｔｏｆｔｒａｉｎｉｎｇｐｒｏｃｅｓｓ㊀㊀训练过程损失如图８所示ꎬ在ＧＡＮ模型训练过程中由于生成网络和判别网络的对抗ꎬ在训练一段时间达到稳定期后ꎬ生成网络和判别网络的损失都应该在一个小区间内波动ꎬ而不会有明显的持续上升或者下降趋势ꎮ在图８中ꎬ前２００个ｅｐｏｃｈ的学习率相对较大ꎬ总损失ｔｏｔａｌ＿ｌｏｓｓ曲线震荡明显ꎬ２００个ｅｐｏｃｈ后随着学习率的降低和训练达到稳定期ꎬｔｏｔａｌ＿ｌｏｓｓ在较小的区间波动ꎮ在４００个ｅｐｏｃｈ时产生的图像已经有了蓝印花布纹样的风格并且网络已稳定ꎮ图８㊀训练过程ｔｏｔａｌ＿ｌｏｓｓ变化Ｆｉｇ.８㊀Ｔｏｔａｌ＿ｌｏｓｓｃｈａｎｇｅｓｄｕｒｉｎｇｔｒａｉｎｉｎｇ３.３㊀网络对比为了验证所提出网络的优越性ꎬ本文将提出的方法与对于同样基于对偶思想的ＣｙｃｌｅＧＡＮ㊁ＤｕａｌＧＡＮ[１６]㊁ＤｉｓｃｏＧＡＮ[１７]进行对比ꎬ结果如图９所示ꎮ在图９中ꎬ从上到下依次为原图㊁ＣｙｃｌｅＧＡＮ算法㊁ＤｕａｌＧＡＮ算法㊁ＤｉｓｃｏＧＡＮ算法与ＳＥ￣ＣｙｃｌｅＧＡＮ(ｏｕｒｓ)算法生成的结果ꎮ对于ＤｕａｌＧＡＮ和ＤｉｓｃｏＧＡＮ来说ꎬ其生成网络采用Ｕｎｅｔ结构ꎬ保留了大量原始的细节ꎬ在灰度图中仍能隐约看到原始的图像ꎬ但是在高度抽象化的蓝印花布风格中ꎬ这些细节反而使得生成效果大大降低ꎬ并且图像中没有产生合适断线ꎬ整体结构也更加混乱ꎮＣｙｃｌｅＧＡＮ虽然表现出断线ꎬ但是断线后形成的色块过于紧凑ꎬ没有呈现蓝印花布的风格ꎮＳＥ￣ＣｙｌｃｅＧＡＮ(ｏｕｒｓ)明显提升了图像的抽象化程度ꎬ相对于其他网络色块少而精炼ꎬ主体更加突出ꎬ线条分布合理ꎬ 线断意连风格明显ꎬ表现最优ꎮ图９㊀不同网络生成结果Ｆｉｇ.９㊀Ｇｅｎｅｒａｔｅｄｒｅｓｕｌｔｓｏｆｄｉｆｆｅｒｅｎｔｎｅｔｗｏｒｋｓ３.４㊀注意力机制对比为了直观展示使用不同注意力机制的风格迁移效果ꎬ本文将ＳＥ注意力机制替换为ＣＢＡＭ注意力机制[１８]和ＥＣＡ注意力机制[１９]并进行对比ꎬ结果如图１０所示ꎮ在图１０中ꎬ从上到下依次为原图㊁ＣｙｃｌｅＧＡＮ算法㊁ＣＢＡＭ￣ＣｙｃｌｅＧＡＮ算法㊁ＥＣＡ￣ＣｙｃｌｅＧＡＮ算法与ＳＥ￣ＣｙｃｌｅＧＡＮ(ｏｕｒｓ)算法生成的结果ꎮ其中ＣＢＡＭ￣ＣｙｃｌｅＧＡＮ对于原图像的细节保留过多ꎬ断线产生的位置不合理ꎬ转化后形成的风格比起ＳＥ￣ＣｙｃｌｅＧＡＮ(ｏｕｒｓ)有较大的不足ꎬ在小猫图像和狐狸图像上较为明显ꎬ且ＣＢＡＭ￣ＣｙｃｌｅＧＡＮ明显产生了更多的噪点ꎮ而ＥＣＡ￣ＣｙｃｌｅＧＡＮ对于原图内容的把握明显缺失ꎬ色块排列杂乱无章ꎬ没有明确的主题或焦点ꎬ难以捕捉到图片想要表达的核心信息ꎬ甚至生成了如第二列蜜蜂图所示的意义不明的图像ꎮＳＥ￣ＣｙｃｌｅＧＡＮ(ｏｕｒｓ)生成的效果最好ꎬ色块更加写意ꎬ色块边缘更加柔和ꎬ如牛的耳朵细看之下宛如花瓣ꎬ并且断线的位置恰到好处ꎬ符合蓝印花布线断意连的风格ꎬ在整体上结构清晰ꎬ有更高的艺术价值ꎮ实验结果表明ꎬ相较于其他算法ꎬ使用ＳＥ注意力机制的ＣｙｃｌｅＧＡＮ算法在生成蓝印花布图样上有明显的优势ꎮ图１０㊀使用不同注意力的改进ＣｙｃｌｅＧＡＮ网络生成结果Ｆｉｇ.１０㊀ＩｍｐｒｏｖｅｄＣｙｃｌｅＧＡＮｎｅｔｗｏｒｋｇｅｎｅｒａｔｅｄｒｅｓｕｌｔｓｕｓｉｎｇｄｉｆｆｅｒｅｎｔａｔｔｅｎｔｉｏｎｍｅｔｈｏｄｓ５３Ｖｏｌ.６１㊀Ｎｏ.１ＳｉｎｇｌｅｐａｔｔｅｒｎａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮ４㊀结㊀论作为国家非物质文化遗产的蓝印花布ꎬ对其进行数字化传承和创新有重要的价值和意义ꎮ本文提出了一种基于ＳＥ注意力机制的改进ＣｙｃｌｅＧＡＮ的端到端网络ＳＥ￣ＣｙｃｌｅＧＡＮꎮ通过ＳＥ注意力模块和残差模块重新塑造的生成网络提高了原始ＣｙｃｌｅＧＡＮ的几何变形能力ꎬ相对于其他网络ꎬ图像的内容与蓝印花布的风格有更好的融合ꎬ产生的噪点也较少ꎬ并且新提出的网络本身有小样本训练ꎬ无需配对数据集的特点ꎬ可以很好地迁移到其他类似纹样的生成ꎮ但是因为数据集和分辨率的限制ꎬ目前生成的纹样仍较为简单ꎮ而且ꎬ受到笔画和图像信息密集程度的影响较大ꎬ难以生成更加复杂㊁内容丰富多样的纹样ꎮ接下来将在多类㊁复杂的蓝印花布纹样自动生成方面开展研究工作ꎮ«丝绸»官网下载㊀中国知网下载参考文献:[１]侯莉莉ꎬ须秋洁.蓝印花布纹样设计之点画技法探析[Ｊ].美术教育研究ꎬ２０２２(１５):５０￣５２.ＨＯＵＬＬꎬＸＵＱＪ.Ａｎａｌｙｓｉｓｏｆｓｔｉｐｐｌｉｎｇｔｅｃｈｎｉｑｕｅｓｉｎｂｌｕｅｃａｌｉｃｏｐａｔｔｅｒｎｄｅｓｉｇｎ[Ｊ].ＡｒｔＥｄｕｃａｔｉｏｎＲｅｓｅａｒｃｈꎬ２０２２(１５):５０￣５２.[２]金晓伟.蓝印花布图案的视觉语言研究[Ｊ].美术观察ꎬ２０２２(８):７３￣７４.ＪＩＮＸＷ.Ｓｔｕｄｙｏｎｔｈｅｖｉｓｕａｌｌａｎｇｕａｇｅｏｆｂｌｕｅｃａｌｉｃｏ[Ｊ].ＡｒｔＯｂｓｅｒｖａｔｉｏｎꎬ２０２２(８):７３￣７４.[３]杜威.蓝印花布艺术传承研究[Ｊ].西部皮革ꎬ２０２２ꎬ４４(２１):５４￣５６.ＤＵＷ.Ｓｔｕｄｙｏｎｔｈｅｉｎｈｅｒｉｔａｎｃｅｏｆｂｌｕｅｃａｌｉｃｏｆａｂｒｉｃａｃｔ[Ｊ].ＷｅｓｔＬｅａｔｈｅｒꎬ２０２２ꎬ４４(２１):５４￣５６.[４]贾小军ꎬ叶利华ꎬ邓洪涛ꎬ等.基于卷积神经网络的蓝印花布纹样基元分类[Ｊ].纺织学报ꎬ２０２０ꎬ４１(１):１１０￣１１７.ＪＩＡＸＪꎬＹＥＬＨꎬＤＥＮＧＨＴꎬｅｔａｌ.Ｅｌｅｍｅｎｔｓｃｌａｓｓｉｆｉｃａｔｉｏｎｏｆｖｅｉｎｐａｔｔｅｒｎｓｕｓｉｎｇｃｏｎｖｏｌｕｔｉｏｎａｌｎｅｕｒａｌｎｅｔｗｏｒｋｓｆｏｒｂｌｕｅｃａｌｉｃｏ[Ｊ].ＪｏｕｒｎａｌｏｆＴｅｘｔｉｌｅＲｅｓｅａｒｃｈꎬ２０２０ꎬ４１(１):１１０￣１１７.[５]贾小军ꎬ邓洪涛ꎬ滕姿ꎬ等.应用轮廓线拟合提取蓝印花布图案基元[Ｊ].纺织学报ꎬ２０１８ꎬ３９(８):１５０￣１５７.ＪＩＡＸＪꎬＤＥＮＧＨＴꎬＴＥＮＧＺꎬｅｔａｌ.Ｅｘｔｒａｃｔｉｏｎｏｆｉｍａｇｅｅｌｅｍｅｎｔｓｆｏｒｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎｃｏｎｔｏｕｒｆｉｔｔｉｎｇ[Ｊ].ＪｏｕｒｎａｌｏｆＴｅｘｔｉｌｅＲｅｓｅａｒｃｈꎬ２０１８ꎬ３９(８):１５０￣１５７.[６]ＧＡＴＹＳＬꎬＥＣＫＥＲＡꎬＢＥＴＨＧＥＭ.Ｉｍａｇｅｓｔｙｌｅｔｒａｎｓｆｅｒｕｓｉｎｇｃｏｎｖｏｌｕｔｉｏｎａｌｎｅｕｒａｌｎｅｔｗｏｒｋｓ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).ＬａｓＶｅｇａｓ:ＩＥＥＥꎬ２０１６.[７]ＷＡＮＧＬꎬＷＡＮＧＺꎬＹＡＮＧＸꎬｅｔａｌ.Ｐｈｏｔｏｇｒａｐｈｉｃｓｔｙｌｅｔｒａｎｓｆｅｒ[Ｊ].ＴｈｅＶｉｓｕａｌＣｏｍｐｕｔｅｒꎬ２０２０ꎬ３６(２):３１７￣３３１.[８]ＬＵＡＮＦꎬＰＡＲＩＳＳꎬＳＨＥＣＨＴＭＡＮＥꎬｅｔａｌ.Ｄｅｅｐｐｈｏｔｏｓｔｙｌｅｔｒａｎｓｆｅｒ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).ＨａｗａｉｉＳｔａｔｅ:ＩＥＥＥꎬ２０１７.[９]ＪＯＨＮＳＯＮＪꎬＡＬＡＨＩＡꎬＬＩＦＦ.Ｐｅｒｃｅｐｔｕａｌｌｏｓｓｅｓｆｏｒｒｅａｌ￣ｔｉｍｅｓｔｙｌｅｔｒａｎｓｆｅｒａｎｄｓｕｐｅｒ￣ｒｅｓｏｌｕｔｉｏｎ[Ｃ]//Ｅｕｒｏｐｅａｎｃｏｎｆｅｒｅｎｃｅｏｎｃｏｍｐｕｔｅｒｖｉｓｉｏｎ(ＥＣＣＶ).Ｃｈａｍ:Ｓｐｒｉｎｇｅｒꎬ２０１６:６９４￣７１１.[１０]ＧＯＯＤＦＥＬＬＯＷＩꎬＰＯＵＧＥＴ￣ＡＢＡＤＩＥＪꎬＭＩＲＺＡＭꎬｅｔａｌ.Ｇｅｎｅｒａｔｉｖｅａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｓ[Ｊ].ＣｏｍｍｕｎｉｃａｔｉｏｎｓｏｆｔｈｅＡＣＭꎬ２０２０ꎬ６３(１１):１３９￣１４４.[１１]ＭＩＲＺＡＭꎬＯＳＩＮＤＥＲＯＳ.Ｃｏｎｄｉｔｉｏｎａｌｇｅｎｅｒａｔｉｖｅａｄｖｅｒｓａｒｉａｌｎｅｔｓ[Ｊ].ＣｏｍｐｕｔｅｒＳｃｉｅｎｃｅꎬ２０１４:２６７２￣２６８０.[１２]ＩＳＯＬＡＰꎬＺＨＵＪＹꎬＺＨＯＵＴꎬｅｔａｌ.Ｉｍａｇｅ￣ｔｏ￣ｉｍａｇｅｔｒａｎｓｌａｔｉｏｎｗｉｔｈｃｏｎｄｉｔｉｏｎａｌａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｓ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).ＨａｗａｉｉＳｔａｔｅ:ＩＥＥＥꎬ２０１７.[１３]ＺＨＵＪＹꎬＰＡＲＫＴꎬＩＳＯＬＡＰꎬｅｔａｌ.Ｕｎｐａｉｒｅｄｉｍａｇｅ￣ｔｏ￣ｉｍａｇｅｔｒａｎｓｌａｔｉｏｎｕｓｉｎｇｃｙｃｌｅ￣ｃｏｎｓｉｓｔｅｎｔａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｓ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＩｎｔｅｒｎａｔｉｏｎａｌＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎ(ＩＣＣＶ).ＨａｗａｉｉＳｔａｔｅ:ＩＥＥＥꎬ２０１７.[１４]ＣＨＥＮＹꎬＬＡＩＹＫꎬＬＩＵＹＪ.Ｃａｒｔｏｏｎｇａｎ:Ｇｅｎｅｒａｔｉｖｅａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｓｆｏｒｐｈｏｔｏｃａｒｔｏｏｎｉｚａｔｉｏｎ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).ＳａｌｔＬａｋｅＣｉｔｙ:ＩＥＥＥꎬ２０１８.[１５]ＨＵＪꎬＳＨＥＮＬꎬＳＵＮＧ.Ｓｑｕｅｅｚｅ￣ａｎｄ￣ｅｘｃｉｔａｔｉｏｎｎｅｔｗｏｒｋｓ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).ＳａｌｔＬａｋｅＣｉｔｙ:ＩＥＥＥꎬ２０１８.[１６]ＹＩＺꎬＺＨＡＮＧＨꎬＴＡＮＰꎬｅｔａｌ.Ｄｕａｌｇａｎ:Ｕｎｓｕｐｅｒｖｉｓｅｄｄｕａｌｌｅａｒｎｉｎｇｆｏｒｉｍａｇｅ￣ｔｏ￣ｉｍａｇｅｔｒａｎｓｌａｔｉｏｎ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＩＥＥＥＩｎｔｅｒｎａｔｉｏｎａｌＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎ(ＩＣＣＶ).ＨａｗａｉｉＳｔａｔｅ:ＩＥＥＥꎬ２０１７.[１７]ＫＩＭＴꎬＣＨＡＭꎬＫＩＭＨꎬｅｔａｌ.Ｌｅａｒｎｉｎｇｔｏｄｉｓｃｏｖｅｒｃｒｏｓｓ￣ｄｏｍａｉｎｒｅｌａｔｉｏｎｓｗｉｔｈｇｅｎｅｒａｔｉｖｅａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｓ[Ｃ]//ＩｎｔｅｒｎａｔｉｏｎａｌＣｏｎｆｅｒｅｎｃｅｏｎＭａｃｈｉｎｅＬｅａｒｎｉｎｇ.Ｓｙｄｎｅｙ:ａｒＸｉｖｅ￣ｐｒｉｎｔｓꎬ２０１７.[１８]ＷＯＯＳꎬＰＡＲＫＪꎬＬＥＥＪＹꎬｅｔａｌ.Ｃｂａｍ:Ｃｏｎｖｏｌｕｔｉｏｎａｌｂｌｏｃｋａｔｔｅｎｔｉｏｎｍｏｄｕｌｅ[Ｃ]//ＰｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅＥｕｒｏｐｅａｎＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎ(ＥＣＣＶ).Ｂｅｒｌｉｎ:ＳｐｒｉｎｇｅｒＳｃｉｅｎｃｅꎬ２０１８.[１９]ＷＡＮＧＱＬꎬＷＵＢＧꎬＺＨＵＰＦꎬｅｔａｌ.ＥＣＡ￣Ｎｅｔ:Ｅｆｆｉｃｉｅｎｔｃｈａｎｎｅｌａｔｔｅｎｔｉｏｎｆｏｒｄｅｅｐｃｏｎｖｏｌｕｔｉｏｎａｌｎｅｕｒａｌｎｅｔｗｏｒｋｓ[Ｃ]//Ｐｒｏｃｅｅｄｉｎｇｓｏｆｔｈｅ２０２０ＩＥＥＥ/ＣＶＦＣｏｎｆｅｒｅｎｃｅｏｎＣｏｍｐｕｔｅｒＶｉｓｉｏｎａｎｄＰａｔｔｅｒｎＲｅｃｏｇｎｉｔｉｏｎ(ＣＶＰＲ).Ｓｅａｔｔｌｅ:ＩＥＥＥꎬ２０２０.６３第６１卷㊀第１期基于ＳＥ注意力ＣｙｃｌｅＧＡＮ的蓝印花布单纹样自动生成ＳｉｎｇｌｅｐａｔｔｅｒｎａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｂａｓｅｄｏｎＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮＲＡＮＥｒｆｅｉ１２ＪＩＡＸｉａｏｊｕｎ１２ＹＵＱｉｎｇｃａｎｇ１ＸＩＥＨａｏ１２ＣＨＥＮＷｅｉｂｉａｏ２３１.ＳｃｈｏｏｌｏｆＣｏｍｐｕｔｅｒＳｃｉｅｎｃｅａｎｄＴｅｃｈｎｏｌｏｇｙＳｃｈｏｏｌｏｆＡｒｔｉｆｉｃｉａｌＩｎｔｅｌｌｉｇｅｎｃｅＺｈｅｊｉａｎｇＳｃｉ￣ＴｅｃｈＵｎｉｖｅｒｓｉｔｙＨａｎｇｚｈｏｕ３１００１８Ｃｈｉｎａ２.ＣｏｌｌｅｇｅｏｆＩｎｆｏｒｍａｔｉｏｎＳｃｉｅｎｃｅａｎｄＥｎｇｉｎｅｅｒｉｎｇＪｉａｘｉｎｇＵｎｉｖｅｒｓｉｔｙＪｉａｘｉｎｇ３１４００１Ｃｈｉｎａ３.ＣｏｌｌｅｇｅｏｆＭａｔｈｅｍａｔｉｃｓａｎｄＣｏｍｐｕｔｅｒＳｃｉｅｎｃｅＺｈｅｊｉａｎｇＮｏｒｍａｌＵｎｉｖｅｒｓｉｔｙＪｉｎｈｕａ３２１００４ＣｈｉｎａＡｂｓｔｒａｃｔＢｌｕｅｃａｌｉｃｏｉｓａｔｒａｄｉｔｉｏｎａｌｃｒａｆｔｐｒｉｎｔｉｎｇａｎｄｄｙｅｉｎｇｐｒｏｄｕｃｔｉｎＣｈｉｎａｗｉｔｈａｌｏｎｇｈｉｓｔｏｒｙ.Ｉｔｉｓａｓｉｍｐｌｅａｎｄｐｒｉｍｉｔｉｖｅｂｌｕｅａｎｄｗｈｉｔｅｆａｂｒｉｃｔｈａｔｒｅｆｌｅｃｔｓｐｅｏｐｌｅｓｐｒｅｆｅｒｅｎｃｅｓａｎｄｅｍｂｏｄｉｅｓｔｈｅｉｒｌｏｎｇｉｎｇｆｏｒａｈａｐｐｙｌｉｆｅａｎｄｓｉｍｐｌｅａｅｓｔｈｅｔｉｃｔａｓｔｅ.Ｉｔｉｓｆａｍｏｕｓｆｏｒｉｔｓｕｎｉｑｕｅｐａｔｔｅｒｎｄｅｓｉｇｎｓｔｙｌｅａｎｄｂｒｏｋｅｎｌｉｎｅｓ.Ｈｏｗｅｖｅｒｔｈｅｌａｃｋｏｆａｎａｌｇｏｒｉｔｈｍｆｏｒｔｈｅａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｏｆｂｌｕｅｃａｌｉｃｏｐａｔｔｅｒｎｓｈａｓｈｉｎｄｅｒｅｄｔｈｅｉｎｎｏｖａｔｉｖｅｒｅｓｅａｒｃｈｏｆｐａｔｔｅｒｎｓ.Ｆｏｒｔｈｉｓｒｅａｓｏｎａｎｅｎｄ￣ｔｏ￣ｅｎｄａｕｔｏｍａｔｉｃｇｅｎｅｒａｔｉｏｎｍｅｔｈｏｄｏｆｔｈｅｂｌｕｅｃａｌｉｃｏｐａｔｔｅｒｎｗａｓｐｒｏｐｏｓｅｄｔｏｒｅａｌｉｚｅｔｈｅａｕｔｏｍａｔｉｃｃｏｎｖｅｒｓｉｏｎｆｒｏｍａｓｉｍｐｌｅｓｔｒｏｋｅｉｍａｇｅｔｏａｓｉｎｇｌｅｂｌｕｅｃａｌｉｃｏｐａｔｔｅｒｎ.ＯｕｒｓｏｌｕｔｉｏｎｂｅｌｏｎｇｓｔｏＣｙｃｌｅＧＡＮ￣ｂａｓｅｄｍｅｔｈｏｄｓｗｈｉｃｈａｒｅｐｏｐｕｌａｒｔｏｓｔｙｌｉｚｅｉｍａｇｅｓｉｎａｒｔｉｓｔｉｃｆｏｒｍｓｓｕｃｈａｓｐａｉｎｔｉｎｇ.ＣｙｃｌｅＧＡＮｉｓａｇｅｎｅｒａｔｉｖｅａｄｖｅｒｓａｒｉａｌｎｅｔｗｏｒｋｂａｓｅｄｏｎｄｕａｌｌｅａｒｎｉｎｇｏｖｅｒｃｏｍｉｎｇｔｈｅｔｈｅｌｉｍｉｔａｔｉｏｎｓｏｆｒｅｑｕｉｒｉｎｇｃｏｒｒｅｓｐｏｎｄｉｎｇｄａｔａｓｅｔｓｏｆｏｔｈｅｒｍｅｔｈｏｄｓｂｕｔｉｔｉｓｓｌｉｇｈｔｌｙｉｎｓｕｆｆｉｃｉｅｎｔｆｏｒｇｅｏｍｅｔｒｉｃｓｈａｐｅｃｈａｎｇｅｓ.ＧｉｖｅｎｔｈｅａｂｓｔｒａｃｔｓｔｙｌｅａｎｄｓｍａｌｌｄａｔａｓｅｔｏｆｂｌｕｅｃａｌｉｃｏｔｈｅｃｏｄｅｒａｎｄｄｅｃｏｄｅｒｉｎｔｈｅＣｙｃｌｅＧＡＮｇｅｎｅｒａｔｉｏｎｎｅｔｗｏｒｋａｒｅｒｅｃｏｎｓｔｒｕｃｔｅｄａｎｄｔｈｅＳＥｓｑｕｅｅｚｅａｎｄｅｘｃｅｐｔｉｏｎａｔｔｅｎｔｉｏｎｍｏｄｕｌｅａｎｄｒｅｓｉｄｕａｌｍｏｄｕｌｅａｒｅｃｏｎｎｅｃｔｅｄｉｎｓｅｒｉｅｓｗｉｔｈｔｈｅｏｒｉｇｉｎａｌｃｏｎｖｏｌｕｔｉｏｎｍｏｄｕｌｅｔｏｉｍｐｒｏｖｅｔｈｅａｂｉｌｉｔｙｏｆｆｅａｔｕｒｅｅｘｔｒａｃｔｉｏｎａｎｄｎｅｔｗｏｒｋｌｅａｒｎｉｎｇ.ＳＥａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｇｉｖｅｓｄｉｆｆｅｒｅｎｔｗｅｉｇｈｔｓｔｏｄｉｆｆｅｒｅｎｔｐｏｓｉｔｉｏｎｓｏｆｉｍａｇｅｓｆｒｏｍｔｈｅｐｅｒｓｐｅｃｔｉｖｅｏｆｔｈｅｃｈａｎｎｅｌｄｏｍａｉｎｗｉｔｈｔｈｅｎｅｔｗｏｒｋｆｏｃｕｓｉｎｇｏｎｋｅｙｉｎｆｏｒｍａｔｉｏｎ.Ａｔｔｈｅｓａｍｅｔｉｍｅｒｅｄｕｃｉｎｇｔｈｅｎｕｍｂｅｒｏｆｒｅｓｉｄｕａｌｂｌｏｃｋｌａｙｅｒｓｏｆｔｈｅｃｏｎｖｅｒｔｅｒｉｎｔｈｅｇｅｎｅｒａｔｉｖｅｎｅｔｗｏｒｋｔｏｒｅｄｕｃｅｏｖｅｒ￣ｆｉｔｔｉｎｇｉｓｉｍｐｌｅｍｅｎｔｅｄ.Ｂｅｓｉｄｅｓｔｏｃａｒｒｙｏｕｔｔｈｉｓｅｘｐｅｒｉｍｅｎｔｗｅｈａｖｅｍａｄｅｔｈｅｄａｔａｓｅｔｏｆａｓｉｎｇｌｅｐａｔｔｅｒｎｏｆｂｌｕｅｃａｌｉｃｏａｎｄｔｈｅｄａｔａｓｅｔｏｆｓｉｍｐｌｅｓｔｒｏｋｅｓ.Ｄｕｅｔｏｔｈｅｌａｃｋｏｆｐｒｅｖｉｏｕｓｒｅｓｅａｒｃｈｏｎｔｈｉｓｉｓｓｕｅａｎｄｔｈｅｌａｃｋｏｆａｐｐｒｏｐｒｉａｔｅｍｅａｓｕｒｅｍｅｎｔｉｎｄｉｃａｔｏｒｓｆｏｒｓｕｃｈｅｘｐｅｒｉｍｅｎｔｓｔｈｅｓｔｕｄｙｕｓｅｄｔｈｅｇｅｎｅｒａｔｅｄｉｍａｇｅｆｏｒｖｉｓｕａｌｃｏｍｐａｒｉｓｏｎ.ＩｎｔｈｅｅｘｐｅｒｉｍｅｎｔｔｏｐｒｏｖｅｔｈｅｓｕｐｅｒｉｏｒｉｔｙｏｆｔｈｅｐｒｏｐｏｓｅｄａｌｇｏｒｉｔｈｍｗｅｃｏｍｐａｒｅｄｏｕｒａｌｇｏｒｉｔｈｍｗｉｔｈｔｈｅｏｒｉｇｉｎａｌＣｙｃｌｅＧＡＮａｎｄｏｔｈｅｒａｌｇｏｒｉｔｈｍｓｂａｓｅｄｏｎｔｈｅｄｕａｌｉｄｅａｌｉｋｅＤｕａｌＧＡＮａｎｄＤｉｓｃｏＧＡＮ.ＴｈｅｅｘｐｅｒｉｍｅｎｔａｌｒｅｓｕｌｔｓｓｈｏｗｔｈａｔｏｕｒｐｒｏｐｏｓｅｄａｌｇｏｒｉｔｈｍｃａｎｅｆｆｅｃｔｉｖｅｌｙｃｏｍｂｉｎｅｔｈｅｃｏｎｔｅｎｔｏｆｓｉｍｐｌｅｓｔｒｏｋｅｓｗｉｔｈｔｈｅｓｔｙｌｅｏｆｂｌｕｅｃａｌｉｃｏｗｈｉｌｅｔｈｅｏｒｉｇｉｎａｌＣｙｃｌｅＧＡＮａｌｇｏｒｉｔｈｍｈａｓｉｎｓｕｆｆｉｃｉｅｎｔｇｅｏｍｅｔｒｉｃｄｅｆｏｒｍａｔｉｏｎａｂｉｌｉｔｙａｎｄｔｈｅｏｔｈｅｒｔｗｏａｌｇｏｒｉｔｈｍｓｓａｖｅｔｏｏｍｕｃｈｏｒｉｇｉｎａｌｉｎｆｏｒｍａｔｉｏｎｗｈｉｃｈｄｏｅｓｎｏｔｃｏｎｆｏｒｍｔｏｔｈｅｃｏｎｃｉｓｅｃｈａｒａｃｔｅｒｉｓｔｉｃｓｏｆｂｌｕｅｃａｌｉｃｏ.ＡｔｔｈｅｓａｍｅｔｉｍｅｔｈｅｓｔｕｄｙａｌｓｏｃｏｍｐａｒｅｄｔｈｅｅｆｆｅｃｔｏｆｕｓｉｎｇｄｉｆｆｅｒｅｎｔａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｓａｎｄｃｏｍｐａｒｅｄｔｈｅＥＣＡｅｆｆｉｃｉｅｎｔｃｈａｎｎｅｌａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｗｉｔｈＣＢＡＭｃｏｎｖｏｌｕｔｉｏｎａｌｂｌｏｃｋａｔｔｅｎｔｉｏｎｍｏｄｕｌｅｍｅｃｈａｎｉｓｍ.ＡｍｏｎｇｖａｒｉｏｕｓａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｓｔｈｅＳＥａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｓｔｉｌｌｈａｓｔｈｅｂｅｓｔｅｆｆｅｃｔ.ＴｈｅｉｍａｇｅｇｅｎｅｒａｔｅｄｂｙｕｓｉｎｇＥＣＡａｔｔｅｎｔｉｏｎｌａｃｋｓｔｈｅｃｏｎｔｅｎｔｏｆｔｈｅｏｒｉｇｉｎａｌｉｍａｇｅｔｈｅｉｍａｇｅｓｔｙｌｅｃｏｎｖｅｒｓｉｏｎｇｅｎｅｒａｔｅｄｂｙｕｓｉｎｇＥＣＡａｔｔｅｎｔｉｏｎｉｓｎｏｔｅｎｏｕｇｈ.ＢｅｓｉｄｅｓｃｏｍｐａｒｅｄｗｉｔｈｔｈｅｏｔｈｅｒｔｗｏｍｅｔｈｏｄｓｔｈｅｉｍａｇｅｇｅｎｅｒａｔｅｄｂｙｕｓｉｎｇｔｈｅＳＥａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｈａｓｓｍｏｏｔｈｅｒｃｏｌｏｒｂｌｏｃｋｅｄｇｅｓａｎｄｌｅｓｓｎｏｉｓｅ.Ｔｈｅｒｅａｓｏｎｆｏｒｔｈｅｇａｐｃａｕｓｅｄｂｙｕｓｉｎｇｄｉｆｆｅｒｅｎｔａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｓｍａｙｂｅｔｈａｔｄｉｆｆｅｒｅｎｔａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｓｉｎｃｒｅａｓｅｔｈｅｌｅａｒｎｉｎｇａｂｉｌｉｔｙｏｆｇｅｎｅｒａｔｏｒｓｔｏｄｉｆｆｅｒｅｎｔｄｅｇｒｅｅｓａｎｄｓｏｍｅｍａｋｅｉｔｄｉｆｆｉｃｕｌｔｆｏｒｇｅｎｅｒａｔｏｒｓａｎｄｄｉｓｃｒｉｍｉｎａｔｏｒｓｔｏｆｏｒｍｅｆｆｅｃｔｉｖｅｃｏｍｐｅｔｉｔｉｏｎ.ＴｈｒｏｕｇｈｔｈｅａｂｏｖｅｅｘｐｅｒｉｍｅｎｔｓｔｈｅｂｌｕｅｃａｌｉｃｏａｕｔｏｍａｔｉｃａｌｌｙｇｅｎｅｒａｔｅｄｂｙｔｈｅｍｅｔｈｏｄｂａｓｅｄｏｎｔｈｅＳＥａｔｔｅｎｔｉｏｎＣｙｃｌｅＧＡＮｎｅｔｗｏｒｋｉｓｃｌｏｓｅｒｔｏｔｈｅｏｒｉｇｉｎａｌｓｔｙｌｅ.Ｔｈｒｏｕｇｈｔｈｉｓａｌｇｏｒｉｔｈｍｔｈｅｄｅｓｉｇｎｐｒｏｃｅｓｓｏｆｔｈｅｂｌｕｅｃａｌｉｃｏｃａｎｂｅｓｉｍｐｌｉｆｉｅｄｗｈｉｃｈｉｓｂｅｎｅｆｉｃｉａｌｔｏｔｈｅｄｉｇｉｔａｌｉｎｈｅｒｉｔａｎｃｅａｎｄｉｎｎｏｖａｔｉｏｎｏｆｔｈｅｂｌｕｅｃａｌｉｃｏ.Ａｓａｎａｔｉｏｎａｌｉｎｔａｎｇｉｂｌｅｃｕｌｔｕｒａｌｈｅｒｉｔａｇｅｔｈｅｂｌｕｅｃａｌｉｃｏｉｓｏｆｉｍｐｏｒｔａｎｔｖａｌｕｅａｎｄｓｉｇｎｉｆｉｃａｎｃｅ.Ｉｎｔｈｉｓｐａｐｅｒａｎｅｎｄ￣ｔｏ￣ｅｎｄｎｅｔｗｏｒｋＳＥ￣ＣｙｃｌｅＧＡＮｂａｓｅｄｏｎｔｈｅＳＥａｔｔｅｎｔｉｏｎｍｅｃｈａｎｉｓｍｉｓｐｒｏｐｏｓｅｄ.Ｃｏｍｐａｒｅｄｗｉｔｈｏｔｈｅｒｎｅｔｗｏｒｋｓｔｈｅｃｏｎｔｅｎｔｏｆｔｈｅｉｍａｇｅｉｓｂｅｔｔｅｒｉｎｔｅｇｒａｔｅｄｗｉｔｈｔｈｅｓｔｙｌｅｏｆｔｈｅｂｌｕｅｃａｌｉｃｏ.Ｔｈｅｐｒｏｐｏｓｅｄｎｅｔｗｏｒｋｄｏｅｓｎｏｔｎｅｅｄｍａｔｃｈｉｎｇｄａｔａｓｅｔｓｓｏｉｔｃａｎｂｅｗｅｌｌｍｉｇｒａｔｅｄｔｏｔｈｅｇｅｎｅｒａｔｉｏｎｏｆｏｔｈｅｒｓｉｍｉｌａｒｐａｔｔｅｒｎｓ.Ｎｅｖｅｒｔｈｅｌｅｓｓｄｕｅｔｏｔｈｅｌｉｍｉｔａｔｉｏｎｏｆｔｈｅｄａｔａｓｅｔａｎｄｒｅｓｏｌｕｔｉｏｎｔｈｅｐａｔｔｅｒｎｓｇｅｎｅｒａｔｅｄａｔｐｒｅｓｅｎｔａｒｅｓｔｉｌｌｒｅｌａｔｉｖｅｌｙｓｉｍｐｌｅ.Ｔｈｅｎｅｘｔｒｅｓｅａｒｃｈｗｏｒｋｗｉｌｌｆｏｃｕｓｏｎｓｔｕｄｙｉｎｇｔｈｅｍｅｔｈｏｄｏｆｇｅｎｅｒａｔｉｎｇｍｕｌｔｉ￣ｃｌａｓｓａｎｄｃｏｍｐｌｅｘｂｌｕｅｃａｌｉｃｏ.ＫｅｙｗｏｒｄｓｂｌｕｅｃａｌｉｃｏＳＥａｔｔｅｎｔｉｏｎｓｔｙｌｅｔｒａｎｓｆｅｒＣｙｃｌｅＧＡＮｓｉｎｇｌｅｐａｔｔｅｒｎｓｅｍｉ￣ｓｕｐｅｒｖｉｓｅｄｌｅａｒｎｉｎｇｉｍａｇｅｇｅｎｅｒａｔｉｏｎ７３。

Autodesk Vault 2011 属性系统简介与概述说明书

AUTODESK® VAULT 2011PROPERTIES INTRODUCTION AND OVERVIEWConcepts and common administrative tasks are described in this paper. This paper is not a comprehensive description - complete details are available through Vault 2011 Help. Most of the features described are in all Vault 2011 products. However, some features are only available in the higher levels of the Vault product line.IntroductionThe property system for Vault 2011 is a single set of properties that are shared across files, items, change orders and reference designators. There are two types of properties: System and User Defined Properties (UDP.) System properties cannot be deleted but do support some configuration options like renaming and a few support mapping. Duplicate property names are not permitted for either type.UDP’s are custom created properties that support assignment to object groups, policy constraints and mapping of values with file and BOM properties. With each new vault there are numerous UDP’s supplied as part of the default configuration.Some of the highlights of the new property system:o Consistent user interface for all property managemento Property constraint overrides by categoryo Streamlined Edit Properties wizardo New vertical properties grid supports multiple files as well as Items & Change Orders o‘Lists’ support text and number data types as well as addition and removal of valueso Standardized mapping for all property sourceso Bi-directional mappingProperty DefinitionA property definition contains a name, data type, policy settings and mappings. The definition also specifies which object groups are associated with and may utilize the property. As an example, we will use the property definition Author. If Author is associated with the File and Item groups it may appear on any file or item but cannot appear with change orders and reference designators. (Reference Designators are a feature of AutoCAD Electrical). Every object (file and item) that is associated with the property definition Author will have a unique value. This may seem obvious when comparing two files as they each may have a unique value. This principle may not be as obvious when comparing objects across groups. If a file is promoted to an item, the file and item are allowed to have unique values for Author.*Change Order Link Properties remain a separate propertysystem.AdministrationCreation and AssociationTo create a property the name and data typemust be specified. The new property is notavailable for use until it has been associated toan object group. The groups are: Change Order,File, Item and Reference Designator. In thesample image below, the File object group isselected. This new property cannot be attachedto an Item, Change Order or ReferenceDesignator unless those object groups are alsoselected.All files in the categories Base or Engineering will have this property automatically attached. If this property needs to be attached to a specific file in another category it may be manually attached. Manual attachment can be done in two ways: using the Edit Properties Wizard or the Add or Remove Property located on the Actions menu.The object groups Change Order and Reference Designator do not support categories. Therefore, any property associated with one of these groups will be automatically attached to all objects in that group.SettingsThe policy values under the Property Valuescolumn (left side of the dialog) are applied toall instances of this property except where thecategory override applies. The CategoryValues allow overrides by category. Consultthe Help for further details about overrides andpolicies. In this paper, we will outline InitialValue, List Values and Enforce List Values.Initial ValueThe Initial Value is applied once when theproperty is initially associated with an object.The initial value is only applied in the absenceof a user supplied or a mapped value.The initial association occurs in three circumstances: 1) object is created (ex: adding a file or creating an item) 2) assignment to a category that automatically attaches the property 3) manual property attachment.There are two types of Initial Value: static and mapped. The static value is a fixed value and may be any value that is valid for the selected data type. An initial mapped value copies the value from a file or BOM property.Initial Values should NOT be used onproperties where all regular mappings read thevalue from a file or BOM. A blank value in themapped file or BOM field takes precedenceover the initial value. This may appear as ifthe initial value is not applied when in fact themapped value of ‘blank’ takes precedence.List ValuesProperties of type Text and Number mayprovide a list of values for user selection andsearching. The administrator may add orremove values from the list at any time.Removal of a value from the list does notremove the value from any property where thatvalue has been applied. When specifying thevalue for this property, the user may chosefrom the list of values. Enter values that arenot on the list is allowed. If this property ismapped to read a value from a file or BOM, the imported value is not required to be on the list. Enforce List ValuesWhen enabled, this option will provide a warning symbol adjacent to this property if the value is not on the list. When a value is in violation of this policy, the default configuration for lifecycle transitions will not allow a file or item to be released.MappingTo create a property mapping, the administrator must first choose which object group is to be mapped. In the image below, this is specified under the first column titled Entity . The available choices are based on the value of the Associations field. Several Content Providers are included but in most cases it is best to leave theselection on All Files (*.*). Vault willautomatically select the most appropriateContent Provider based on the file type.Next, select a file that contains the propertyor BOM field to be mapped. The image onthe left shows the file properties available formapping in the file manifold_block.ipt .The Type column shows the data type of thesource property. Mapping may be doneacross data types. However, there arespecial considerations that are detailed in thenext section. The mapping direction bydefault will chose bi-directional unless the fileor BOM property does not support the inputof values. When this occurs the mappingoption will be limited to Read only. Readonly mappings should be used sparinglybecause any UDP that contains only ‘Readonly’ mappings may not be modified in Vault.Mapping Across Data TypesThere are four property types: Text,Number, Boolean & Date . The following matrix defines valid property mappings. Whenever a mapping is created between two different property types there is the possibility of incompatibility. The onus is on the user to input valid values. If an invalid value is entered in most cases, the equivalence will flag the property as non-equivalent. The exceptions are listed below.1. Mapping Boolean with Text : The supported valid text values are: Yes/No , True/False and1/0. These values are localized. A string like ‘Autodesk’ entered in a Text property cannot be transferred to a Boolean property. This property mapping would be flagged as notequivalent.2. Mapping Text with Number or Text with Date : Works well when all clients and the serverare in the same language-locale. With mixed locales values may convert in a manner that is not intuitive and may produce an undesirable result. Therefore, mapping Text withNumber or Text with Date is only recommended when the server and all clients are working in the same locale.Create OptionThe Create option applies to write mappings; if the file property does not exist when a value is pushed to the file, the administrator may choose whether the file property is created or not. The Create option has another function that is not obvious: when enabled the equivalence calculation will consider the absence of the property definition in the file as a blank value and Supported mapping across data types Source Property (File or BOM) U DP Text Number Boolean Date Text Yes Yes (2) Yes (1) Yes (2) Number Yes (2) Yes Yes NoBoolean Yes (1) Yes Yes No Date Yes (2) No No Yescompare it against the value of the UDP in Vault. When the Create option is disabled, equivalence will be set to ‘Good’ when the mapped property definition does not exist in the file.Example: I have two departments in myorganization that both create .dwg files but theyuse different file properties to represent thesame information. The R&D department usesthe file property DwgNum. The Toolingdepartment uses the file property DrwNo. I wantto manage all drawings from both groups in asingle Vault and with one UDP ‘DrawingNumber’. The correct configuration is to createbidirectional mappings and set the Create optionto Off for both mappings. The result is that amodification of the UDP Drawing Number willwrite its value back to whichever property existsand it will not create an extra property.Mapping AutoCAD Block AttributesAutodesk® AutoCAD® block attribute mapping requires configuration on the ADMS. Select Index Block Attributes… from the Tools menu in Autodesk Data Management Server Console 2011. Enter the AutoCAD block names from which to extract attributes. After this is done, it is possible to map a UDP to an attribute using the mapping processdescribed above. Configured mappings allow thesystem to read and/or write values between the UDPand the attribute.Usage of attribute mapping is intended for singleinstances of a block or when all block instances havethe same attribute values. It is not possible for multipleblock instances to be mapped to separate UDP’s. Manycompanies have one instance of a title block in a given.dwg files. Occasionally, there are companies that use multiple instances of a title block in a single file. In these cases, the attributes often share the same values. An example is a drawing file that contains three borders of different size. Each border uses the same title block with attributes. The attributes for Customer Name, Engineer, Project Number, etc. will share the same value for all instances. Such attributes that share the same value may be mapped to a UDP. Attributes like Border Size will have a unique value for each block instance. Therefore, Border Size should not be mapped to a UDP in Vault.AutoCAD MechanicalAutodesk® AutoCAD® Mechanical software (ACM) supports three distinct sets of properties, all ofwhich may be mapped to Vault UDPs. The three ACM property sets are: file, assembly and component. See the ACM documentation for details about the intended use and differences between these properties.Vault file properties may map to ACM file properties and Vault item properties may map to ACM assembly and component properties.It should also be noted that ACM assembly and file properties having the same name, should not be mapped to the same Vault UDP.AutoCAD ElectricalAutodesk® AutoCAD® Electrical software (ACE) supports both file and BOM properties. ACE BOM properties may be mapped to Item properties. ACE utilizes properties located in .dwg’s,.wdp’s and associated databases. ACE properties are exposed to Vault in four ways:First: Ordinary DWG™ file properties and block attributes may be mapped to Vault File objects. The majority of these mappings support bi-directional mapping. Creation of these mappings is described in the Mapping section of this document.Second: WDP properties support mapping to Item properties. They also support bi-directional mapping. Creating a mapping with WDP properties requires the AutoCAD Electrical Content Source Provider. The provider isspecified in the second columnof the image at the right. Thisprovider is automatically setwhen a file of type .wdp isselected under the File Propertycolumn. If an associated .wdlfile has been created both theline number and the alternateproperty name will automaticallyappear in the list for selection.You may select the line numberor the alternate display name tocreate the mapping. All wdlproperties will appear in the listof selectable properties; it does not matter if a value is present.Third: Component BOM properties may be mapped to Item properties. This includes properties like:Catalog Number, Component Name, Component Type, Electrical Type, Equivalence Value & Manufacturer and more...To create a mapping to a component BOM property, create a new UDP and associate it to Items. Then on the Mapping tab create a new mapping, making sure the first column Entity is set to Item. Under the File Property column, browse and select any file that contains the property to which you will create the mapping. Some properties require that a value exist or the property is not available for selection in the list.Reminder: When creating new properties it is best to associate them to a category which will automatically associate them to the files and/or items where the property should appear. If this is not done, the property will have to be manually associated to the file or item.Fourth: Reference Designator properties, when mapped will appear in Vault as optional data on an Item BOM. There are eighteen Reference Designator properties available:INST, LOC, TAG, DESC1...3, RATING1 (12)These properties may be mapped to an Item BOM using the DWG content source provider.To create a mapping to a Reference Designator, create a new UDP and associate it to Reference Designator. Then on the Mapping tab create a new mapping, ensure the first column Entity is set to Reference Designator. Under the File Property column select the dwg containing the Reference Designator to which the mapping needs to be created. All Reference Designators are available for selection in the list without requiring a value.Properties(Historical)A handful of properties have duplicates having the same display name with ‘(Historical)’ appended to the end: State, Revision Scheme, Property Compliance, Lifecycle Definition, Category Name & Category Glyph. These ‘historical’ properties exist solely to retain a record of the values when a configuration change alters the value of the non-historical properties. In other words, the ‘historical’property will always contain the value as it existed when that version was created. This situation arises because these properties may have a new value due to a configuration change, even though a new version is not created.A policy change is a good example of why these historical’ properties exist. An organization may have released documents that use the property Vendor. Currently the policy on the property Vendor does not require a value. The administrator modifies the policy ‘Requires Value’ to require a value. After the automatic compliance recalculation, any existing documents (including released documents) with the Vendor property and without a value will have a new PropertyCompliance value of non-compliant. PropertyCompliance(Historical) will retain the value of compliant. MigrationThe property features of Vault 2011 are a significant enhancement. A feature overhaul of this scale poses challenges for migration. Most prominent is the calculation of property compliance. In some migration cases, the compliance calculation will require additional information beyond that which was stored in Vault 2010 or earlier versions. Performing a Re-Index will resolve the majority of these cases. It is highly recommended that a Re-Index is performed after migration. A Re-Index using the option for Latest and Released Versions Only is sufficient. In rare cases, a re-index may not restore compliance values to pre-migration values. If this occurs, manual adjustment to the property configuration may be required.File Index PropertiesFIP’s are no longer supported. The values contained by the FIP’s will remain available in UDP’s. There are multiple FIP configurations that require unique migration rules, listed here:FIP with no mapping or grouping: this ordinary FIP exists in Vault 2010 or earlier, without any mapping to a UDP and is not a member in any group. Migration will create a UDP, which will be mapped to the file property from which the FIP was created.FIP mapped to a UDP: upon migration, the UDP is carried forward and the FIP is removed from Vault. The value remains available in Vault through the UDP.Grouped FIP’s: property groups are migrated to a UDP having the same name and are mapped to the sources of all the grouped FIP’s.Bi-directional MappingsNew to Vault 2011 is the ability to create Bi-directional property mapping. In previous releases, a mapping was either Read or Write. Because of this change, a UDP that has only Read mappings may not be modified. An example is a UDP that is mapped to Read its value from the file property Creation Date. It makes no sense to write a value back to Creation Date.After migrating to Vault 2011, property mappings that were previously Read will be changed to Bi-directional. If the mapped source does not support input of a value, like the Creation Date example above, the mapping will not be changed and will remain Read. UDP’s that have multiple mappingsthrough the same Content Provider may, under specific circumstances, become non-compliant. If this occurs, it may be necessary to alter the configuration to restore compliance.An example:Vault 2010 or any previous version has a property configuration where two or more fileproperties are mapped as Read into the same UDP. This can occur when companiesmerge or when file property name standards change. For the Read mappings of theconfiguration below, equivalence is calculated on the highest priority mapping, which isEng; the mappings to the other properties are ignored.Upon migration to Vault 2011, Read mappings are converted to Bi-directional (shown below.). For the Bi-directional mappings of the configuration below, equivalence iscalculated between the UDP and each file property that exists in the file. In most cases, only one of the file properties exists in any given file, which will result in the UDP being flagged as compliant.If two properties exist in a file both will be considered for equivalence. If either file property has a value that does not match the UDP it is flagged as non-compliant.Enabling the Create option on a mapping will force equivalence calculation on that mapping even when the property definition does not exist in the file. When the property definition does not exist in the file, each mapping with the Create option set to Off is ignored for equivalence calculation.Autodesk, AutoCAD, and DWG are either registered trademarks ortrademarks of Autodesk, Inc., in the USA and/or other countries. All otherbrand names, product names, or trademarks belong to their respectiveholders. Autodesk reserves the right to alter product offerings andspecifications at any time without notice, and is not responsible fortypographical or graphical errors that may appear in this document.© 2010 Autodesk, Inc. All rights reserved.。

spThin包的中文名字：空间细化功能的R包说明书

Package‘spThin’October14,2022Type PackageTitle Functions for Spatial Thinning of Species Occurrence Records for Use in Ecological ModelsVersion0.2.0Date2019-11-14Description A set of functions that can be used to spatially thinspecies occurrence data.The resulting thinned data can be used in ecological modeling,such as ecological niche modeling.BugReports https:///mlammens/spThin/issuesDepends spam,grid,ﬁelds,knitrImports grDevices,graphics,utilsLazyData TRUELicense GPL-3VignetteBuilder knitrRoxygenNote7.0.0NeedsCompilation noAuthor Matthew E.Aiello-Lammens[aut,cre],Robert A.Boria[aut],Aleksandar Radosavljevic[aut],Bruno Vilela[aut],Robert P.Anderson[aut],Robert Bjornson[ctb],Steve Weston[ctb]Maintainer Matthew E.Aiello-Lammens<**********************> Repository CRANDate/Publication2019-11-1518:10:03UTC12plotThin R topics documented:Heteromys_anomalus_South_America (2)plotThin (2)summaryThin (3)thin (4)thin.algorithm (5)Index6Heteromys_anomalus_South_AmericaOccurrence record locations for Heteromys anomalusDescriptionA dataset containing compiled occurrence record locations for Heteromys anomalus in northerncoastal South America.These records have been examined to check for accurate species identiﬁca-tion.FormatA data frame with201rows and4variablesDetails•SPEC.species name assigned to occurrence record•LAT.decimal degree latitude value•LONG.decimal degree longitude value•REGION.region,or island,of occurrenceplotThin Plot diagnosis for results of thin functionDescriptionThree plots(selected by which)are currently available:a plot of the number of repetitions versus the number of maximum records retained at each repetition([1]observed values;[2]log transformed) and a histogram of the maximun records retained[3].UsageplotThin(thinned,which=c(1:3),ask=prod(par("mfcol"))<length(which)&&dev.interactive(),...)summaryThin3 Argumentsthinned A list of data.frames returned by thin function.which if a subset of the plots is required,specify a subset of the numbers1:3.ask logical;if TRUE,the user is asked before each plot,see par(ask=.)....other parameters to be passed through to plotting functions.See Alsothin.algorithmthinsummaryThin Summary method for results of thin functionDescriptionSummarize the results of thin function.UsagesummaryThin(thinned,show=TRUE)Argumentsthinned A list of data.frames returned by thin function.show logical;if TRUE,the summary values are printed at the console.ValueReturns a list with the(1)maximun number of records,(2)number of data frames with maximun number of records and(3)a table with the number of data frames per number of records.See Alsothin.algorithmthin4thin thin Spatially thin species occurence dataDescriptionthin returns spatially thinned species occurence data sets.A randomizaiton algorithm(thin.algorithm) is used to create data set in which all occurnece locations are at least thin.par distance apart.Spa-tial thinning helps to reduce the effect of uneven,or biased,species occurence collections on spatial model outcomes.Usagethin(loc.data,lat.col="LAT",long.col="LONG",spec.col="SPEC",thin.par,reps,locs.thinned.list.return=FALSE,write.files=TRUE,max.files=5,out.dir,out.base="thinned_data",write.log.file=TRUE,log.file="spatial_thin_log.txt",verbose=TRUE)Argumentsloc.data A data.frame of occurence locations.It can include several columnns,but mustinclude at minimum a column of latitude values,a column of longitude values,and a column of species names.lat.col Name of column of latitude values.Caps sensitive.long.col Name of column of longitude values.Caps sensitive.spec.col Name of column of species name.Caps sensitive.thin.par Thinning parameter-the distance(in kilometers)that you want records to beseparated by.reps The number of times to repete the thinning process.Given the random processof removing nearest-neighbors there should be’rep’number of different sets ofcoordinates.locs.thinned.list.returnTRUE/FALSE-If true,the‘list‘of the data.frame of thinned locs resulting fromeach replication is returned(see Returns below).thin.algorithm5write.files TRUE/FALSE-If true,new*.csvﬁles will be written with the thinned locs data max.files The maximum number of*csvﬁles to be written based on the thinned data out.dir Directory to write new*csvﬁles toout.base Aﬁle basename to give to the thinned datasets createdwrite.log.file TRUE/FALSE create/append logﬁle of thinning runlog.file Text logﬁleverbose TRUE/FALSE-If true,running details of the function are print at the console.Valuelocs.thinned.dfs A list of data.frames,each data.frame the spatially thinned locations of the algo-rithm for a single replication.This list will have‘reps‘elements.See Alsothin.algorithmthin.algorithm Implements random spatial thinning algorithmDescriptionthin.algorithm implements a randomization approach to spatially thinning species occurence data.This function is the algorithm underlying the thin function.Usagethin.algorithm(rec.df.orig,thin.par,reps)Argumentsrec.df.orig A data frame of long/lat points for each presence record.The data.frame should be a two-column data frame,one column of long and one of lat thin.par Thinning parameter-the distance(in kilometers)that you want records to be separated by.reps The number of times to repete the thinning process.Given the random process of removing nearest-neighbors there should be’rep’number of different sets ofcoordinates.Valuereduced.rec.dfs:A list object of length’rep’.Each list element is a different data.frame of spatially thinned presence records.Index∗datasetsHeteromys_anomalus_South_America,2Heteromys_anomalus_South_America,2 plotThin,2summaryThin,3thin,3,4,5thin.algorithm,3–5,56。

TA Instruments TRIOS软件简介说明书

TRIOS Software OverviewTRIOS is TA Instruments’ state-of-the-art software package that uses cutting-edge technology for instrument control, data collection, and data analysis for thermal analysis and rheology instruments. The intuitive user interface allows you to simply and effectively program experiments and move easily between processing experiments and viewing and analyzing data. TRIOS software delivers a whole new experiment experience.•Easy organization and data fi le management •A unique fi le-naming system allows for effortless organization of data fi les •The History View and File Manager offer simple data fi le location •Compatibility with the latest Windows Operating System platforms •M aximum fl exibility •Instrument control and data analysis via any networked computer •Confi gurable for multiple monitors •Ability to control multiple instruments at once •Remote data analysis•Seamless integration between instrument control and data analysis•Easy data export in a variety of output formats, including XML, Word, Excel, and PDF •Simple graph formatting using The Ribbon •Customization of the displayThe TRIOS software is supported by a full range of services, including onsite training, customer service that is only a phone call away, and easy-to-use, easy-to-understand online help. All of these items refl ect TA Instruments’ commitment to providing thermal analysis and rheology products and related services that deliver maximum value for your investment.TA InstrumentsWhat’s New in TRIOS SoftwareWhat’s New in TRIOS Software V3.1TRIOS software is now better than ever with increased stabilization and key fi xes and enhancements, including User Interface changes that will make your TRIOS experience exceptional. The next generation of instrument control and data collection and analysis, TRIOS V3.1 is more effi cient and intuitive, allowing you to work faster and easier.General TRIOS EnhancementsLimited-Bandwidth Download OptionYou can now download TRIOS faster than ever by installing a version of TRIOS that does not include TRIOS Online Help.Ribbon ChangesThe Ribbon has been simplifi ed with the Home and View tabs consolidated into a single Experiment tab, removing redundancy and unused functionality.Previous Home tab:The File Manager button no longer exists. The ability to close the File Manager was removed, and overlay and analysis document creation is now only available from the File M anager right-click menu. New analysis is available from the Analysis tab. The Geometry information is now available from the Experiment tab. When a geometry is attached to the instrument, you can select the installed geometry from the list of geometry fi les previously created on the system. If your geometry does not appear in the list, clicking Add New Geometry will launch the New Geometry wizard. Editing a geometry can now only be done from the File Manager’s Geometries pane or on the geometry Experiment node.Previous View tab:Document Views functionality (creating a new spreadsheet and graph) as well as Layout functionality (saving and loading fi les) is now located solely in the File Manager. Switch Documents and Views functionality was removed. Access to the Properties panel now only exists from the right-click menu.New Legend User InterfaceUse the Legend menu to make your Legend customizations in one place. With one click of the mouse, you can select the desired Entry Type, choose what items you want to appear in the legend, modify text color and title justifi cation, and turn on/off the Legend title.Additionally, when editing the Legend directly from the graph, use the Quick Format option for editing text that automatically displays when you select an item in the Legend. From the Quick Format box, you canchange the font face, size, and style.Previous Home tabPrevious View tabNew Legend User InterfaceQuick Format optionQuick Format optionNew Curves User InterfaceFormatting curves on a graph is now easier and more intuitive with the introduction of the Curves Format dialog box.•Choose to format your curves Automatically or Manually. Use the Automatic Formatting option to apply your customizations based on the fi le, step, and/or variable, or use the Manual Formatting option to apply your customizations per curve. •Use the palette to specify the order of the colors, symbols, line styles, and extra symbols used on your curves. •Additional formatting options include setting the line thickness, symbol size, and symbol density, and adding extra symbols to the curve •Set the style selection with the use of Quick Styles so that you can easily and quickly apply previously defi ned formatting to your curvesRHEOLOGYARES-G2 EnhancementsOrthogonal Superposition (OSP) FeatureThe Orthogonal Superposition (OSP) feature was added for the ARES-G2. In the OSP mode, the normal force transducer operates as an actuator applying a small sinusoidal linear deformation to the sample while recording the force at the same time. Instead of holding the transducer shaft at a fi xed position, the shaft can now be periodically oscillated in the vertical direction at small amplitudes.The major applications of the OSP mode include: •The superposition of a small strain oscillatory deformation normal to the direction of steady shear fl ow (Orthogonal Superposition) •Oscillation testing in the two orthogonal directions at the same frequency (2D-SAOS)A new group of test modes has been created for Orthogonal oscillation. These test modes are only available for the Orthogonal double wall concentric cylinder, Parallel plate, and Orthogonal torsion fi xture. This feature requires an ARES-G2 with Serial Number 4010-0383 or higher or an earlier version that has been upgraded together with appropriate geometries.DMA ModeThe ARES-G2 DM A feature is designed to allow geometries such as tension/compression and bending to be used with the instrument. DM A testing uses the standard oscillation test modes; when one of these geometries (Three point bending, Mixed bending, Clamped bending, or Linear tension) is selected, the mode of deformation is changed from shear to linear, with the appropriate set of variables. This feature is limited to oscillation tests only since that is the only motor mode available.This feature requires an ARES-G2 with Serial Number 4010-0383 or higher or an earlier version that has been upgraded together withappropriate geometries.Curves User InterfaceOrthogonal Superposition (OSP) FeatureProportional Axial Force ControlAxial force control on an ARES-G2 now has the ability to adjust the commanded force level to follow changes in sample stiffness. Typically this is used to decrease the axial force on a sample as it softens to avoid issues with samples being squashed or stretched too much as they soften. This is done by taking a reference value for the sample stiffness at the start of the test, and using the ratio of the reference stiffness and the current sample stiffness to adjust the commanded axial force.This option is enabled by selecting Compensate for stiffness changes. The compensation scaling factor is a value between 0.0 and 1.0 which controls the relationship between the changes in stiffness and the axial force changes. A value of 0.0 results in a constant force, and a value of 1.0 result in the axial force being a linear function of the stiffness ratio.Motor Control PanelThe ARES-G2 motor control panel and Real time variable signal list were updated to allow for better control of sample displacement and strain by using relative rather than absolute positioning.There are now separate signals for motor position (angular offset relative to encoder home position), and displacement (a running change in displacement since a tare in the DSP). The reported strain signal is now based on the geometry strain constant and the measured displacement.Button changes:•Zerodisplacement: Used to zero the displacement signal •Go to home position: Goes to the encoder zero position•Move to orientation angle: M oves to the geometry alignment angle. Enabled only if the active geometry has an alignment angle (i.e. ARES-G2 DMA fi xture)NOTE: The Move to orientation angle has been added to allow thecorrect positioning of the bending and the tension fi xtures for DM A testing after installing the geometry. The orientation angle is calibrated for every geometry and stored with the geometry parameters.ARES-G2 Phase CalibrationThe phase angle calibration code has been updated to use a hermite spline rather than a polynomial fi t to the phase error. This new phase angle correction provides a better fi t at lower frequencies than the polynomial fi t.When the analysis code is invoked, it checks the instrument fi rmware to see if the hermite spline correction is supported. If supported, it uses the new analysis. If this correction is not supported, it reverts back to the older polynomial implementation.The new form shows the user-adjustable number of terms used in the spline, as well as the correction coeffi cients between the measured data and the corrected data for both the phase and amplitude corrections. It also shows the agreement between the calibration data and the spline coeffi cients that are currently loaded in the instrument. An overlay can be created to show this graphically, if desired, by looking at the “Source data” and “Interpolated data” zones.Motor Control PanelARES-G2 Phase Calibration Compensate for stiffness changesARES-G2 Procedure ChangeIn order to repeat the same step in a procedure, the step has to be entered multiple times. With the new repeat function, a single step can be repeated multiple times. In addition some key test parameters can be changed during the repeat runs; for example the temperature can be incremented by a certain value at each repeat run.ARES-G2 Motor BoostTRIOS V3.1 now supports a second power amplifi er to double the motor power output. T his feature allows applying larger strains on high viscosity materials such as rubber compounds. No user interface changes are associated with this feature. TRIOS automatically recognizes when a second power amplifi er is connected and confi gures itself.Electro Rheology Conditioning BlockThe electro rheology conditioning block was updated to allow for voltage ramps and disconnection of the power amplifi er for a better “zero voltage” point.Three different types of voltage ramps are available: •A simple ramp from an initial voltage from a fi nal voltage •A ramp and hold profi le where the voltage is ramped from an initial value to a fi nal value, and the fi nal value is maintained for specifi ed period of time •A double ramp where the voltage is ramped from an initial value to a fi nal value, then back to the initial value The zero voltage mode sets what happens when a zero voltage level is commanded: Command zero volts uses the function generator to commanded a zero volt level into the power amplifi er, Disable amplifi er uses the ARES-G2 enable relay to open the HV circuit.ARES-G2 and RSA-G2 Enhancements Geometry CalibrationsARES-G2/RSA-G2 geometry calibrations were updated to match the method used by DHR/AR rheometers. Previously, geometry-specifi c calibrations where performed by using the Calibration pane in the TRIOS File M anager, and then applying the resulting calibration parameters to the active geometry. The update includes a Calibrations tab that was added to the geometry document, which shows the current calibration value and the date the calibration was performed, as well as hosts a control panel that can be used to run the geometry calibration in place.The Calibration panels displayed depend on the specifi cs of thegeometry.ARES-G2 Procedure ChangeGeometry CalibrationsARES-G2 Procedure ChangeARES-G2 and DHR EnhancementsTribo-Rheometry Accessory for ARES-G2 and DHRThe Tribo-rheometry option for the ARES-G2 and DHR is now available. Tribology test procedures are used to measure the friction coeffi cient, CoF, as a function of the sliding speed under dry and lubricated conditions between any two substrates in contact. Applications range from biological, personal products like creams and lotions, to automotive components and lubrication in machinery design.Tribology tests can only be used with the special Tribo-rheometry geometries. The tribo-rheometry geometry is supported in the following test modes:•Flow Sweep, which is converted to a “Tribology Sweep” when executed.•Transient fl ow steps (Step Rate, Flow Rate, Flow Temperature ramp), which are converted to Tribology Steps when executing.Procedure templates for common Tribology tests can be loaded from the template folder.The modular Tribo-Rheometry Accessory can be confi gured with 4 different geometries (Ring on Plate, Ball on Three Plates, Three Balls on Plate, and Ball on Three Balls), offering a range of contact profi les that are compatible with ARES-G2 FCO and APS or the DHR ETC and Stepped Disposable Peltier Plate.DHR/AR Enhancements•Support for DHR Optics Plate Accessory, Building M aterial Cell, and Bayonet Peltier Plate•DHR Pressure Cell•Support for vane and starch rotor added•Calibration page correctly refl ects mapping status•Resetting of geometry gap fi xed•M ap only applied during test to avoid over speed error when magnetic coupling not engaged•DHR Gap Compensation Calibration: Now allows calibrations to be performed from high to low temperatures•DHR zero gap: Deceleration added to existing standard and axial force modes•DHR Flow Sweep: Scaled time average option fi xed•Concentric cylinder•The end effect fi eld can be changed in a results fi le which will force a recalculation of the data•Rheology Advantage fi les now load with the end effect set to 1•Rheology Advantage fi les measured with ver.1 Double Gap load with the correct immersed heightRheology Analysis•Spriggs and Oldroyd models for oscillation data that were previously available in Rheology Advantage are now available in TRIOSV3.1•Carreau-Yasuda model for fl ow data added•Arrhenius model improved with better starting conditionsTemplate folderNew geometryRheology AnalysisProcedure templateTHERMAL ANALYSISDiscovery TGA EnhancementsModulated TGA (MTGA) SupportM odulated TGA (M TGA) is now supported by TRIOS software V3.1. This option, used with the Discovery TGA, is used to study the same decomposition or volatilization transitions as conventional TGA, plus provides new information that permits unique insights into the behavior of the weight loss reaction — specifi cally, obtaining kinetic information about one or more weight losses, in a shorter period of time than the multiple heating rate approach. M TGA also provides continuous measured values for activation energy throughout the weight loss reaction, not just at specifi c reaction levels.M odulated TGA experiments can be run as either Standard or High Resolution procedures.Discovery DSC Enhancements•To ensure the safety of the user, the Gas 1 selection for the Discovery DSC can now be used for Nitrogen only.INSTALLING TRIOS SOFTW AREFor instructions on installing TRIOS software, refer to the Installing TRIOS Software instructions.ADDITIONAL RESOURCESA number of additional resources are available to you. For assistance with the TRIOS software, fi rst consult the Online Help.For immediate assistance contact the TA Instruments Hotline at +1 302-427-4000 from 8:00 am to 4:30 pm EST.For email support, please send your question to one of the following:t*********************************************************************************************************PREVIOUS WHAT’S NEW DOCUMENTSFor Previous What’s New in TRIOS Software documents, click here.TA INSTRUMENTS OFFICESFor information on our latest products, contact information, and more, see our web site at:TA Instruments — Waters LLCCorporate Headquarters159 Lukens DriveNew Castle, DE 19720USATelephone: 302-427-4000Fax: 302-427-4001Email: **********************。

SAS Studio和SAS Enterprise Guide：哪个是最适合我的SAS编程界面？说明

Paper SAS3044-2019SAS® Studio or SAS® Enterprise Guide®: What's the Best SASProgramming Interface for Me?Amy Peters and Samantha DuPont, SAS Institute Inc.ABSTRACTIn som e people, loyalty to an interface can run as deep as loyalty to a football team, and it’s been no different for SAS® users. The good news is that a concerted effort is underway to rem ove the barriers between interfaces and the differences between them so that you can just choose what you use based on where you are. Are you on a desktop? A m obile device? In a conference room? This paper talks you through the integrated future that’s under developm ent. In the m eantim e, this paper helps you to understand what the strengths and weaknesses are of the program m ing interfaces today so that you can choose wisely. You also learn tips and tricks for working efficiently. The paper focuses on SAS® Studio and SAS® Enterprise Guide®, but it also touches on several of the other SAS program m ing interfaces that are available, including Jupyter.INTRODUCTIONWhen you need to handle an em ail or take som e notes, it’s rare that you give a thought to what app you’ll use. Usually, you just use what happens to be handy on the device you’re on. You take it for granted that your work is coordinated across the devices you use. The sam e should be true for your SAS coding work. And while there’s been good progress, there’s still m ore work to be done before you can seam lessly m ove between the SAS program m ing interfaces currently known as SAS Enterprise Guide and SAS Studio. The goal is for you do your work where it’s m ost convenient for you at the tim e, and as you m ove around it’s easy to pick up where you left off.Figure 1: Create Code in SAS Enterprise Guide, Run it Offline, Access Same Code in SAS StudioWhen this goal is reached, papers like this will be unnecessary. For now, though, there are differences you need to know about to help you choose which interface is best for your work. And there are features you m ight not be aware of that could m ake you m ore efficient. Since things change so quickly, m any of the assets in this paper point to living docum ents on the web so that you can see the current state.HOW DO I CHOOSE WHERE TO WORK?Let’s start with the big picture then drill down to capabilities and specific functions.DO I EVEN HAVE SAS ENTERPRISE GUIDE? SAS STUDIO?There are a few cases where the choice is taken out of your hands:•On a SAS release older than SAS® 9.4 – SAS Enterprise Guide is your only choice as SAS Studio was introduced in SAS 9.4.•Not licensed – While m ost SAS server licenses include unlim ited seats for SAS Enterprise Guide, there are som e in which only a specific num ber are licensed. SAS Studio, however, is free with any SAS license; there is no per seat charge.•Not installed – If it’s a case of an IT policy forbidding desktop applications, then you won’t be able to use SAS Enterprise Guide. Otherwise, it’s a m atter of finding asym pathetic adm inistrator to install SAS Enterprise Guide or to configure SAS Studio and provide you the URL to access it.•On SAS® Viya® – SAS Studio is currently the only native SAS Viya application. You can use both SAS Studio and SAS Enterprise Guide to access a CAS server (subm itcode, get results, load data, and so on) from a SAS 9.4 environm ent. A release ofSAS Enterprise Guide that is native to SAS Viya is planned.For the m ajority of users, both SAS Enterprise Guide and SAS Studio are available. Let’s begin to look at what sweeping capabilities are in one versus the other.WHAT ARE THE MAJOR DIFFERENCES IN CAPABILITY?You can access an interactive version of this content at/software/products/sas-studio/faq/SASStudio_vsEG.htm, which we’ll keep up to date as things progress.ProjectsThis one is easy. If it’s im portant to have your work organized into a projectfor you, then use SAS Enterprise Guide. Projects do not yet exist in SASStudio. If you have existing SAS Enterprise Guide projects (EGP files), youcan read them into SAS Studio 3.8 and they will be converted (using SASStudio process flows, code, and so on). This is a one-way street currently.The future goal is a com m on project form at shared between theapplications.Query ToolA query tool, basically a point-and-click way of building SQL, has been in SAS Enterprise Guide for a very long tim e, so it’s very robust. The query tool in SAS Studio 3.8 is m uch younger, so not as full-featured (for exam ple, it does not currently support creating calculated colum ns). The query tool being built for SAS Studio 5.x is getting a lot closer to the one in SAS Enterprise Guide in term s of what it will be able to do.Process FlowsSim ilar situation to the query tool. SAS Enterprise Guide process flows have m ore features than the process flows in SAS Studio 3.8. More robust flows are com ing for SAS Studio 5.x and, again, the goal is to share a com m on flow.Code CentricitySAS Enterprise Guide was introduced as a point-and-click tool for SAS. SASprogram m ers have been writing code in SAS Enterprise Guide for years andsom etim es com plain that SAS Enterprise Guide hides code or does a bit m oreon their behalf than they’d like. SAS Studio was designed for coding in SAS. Specifically, it was designed to appeal to users fam iliar with the SASWindowing Environm ent (also known as SAS Display Manager or PC SAS).The code editor is front and center and anytim e you use an assistive functionlike a task or query tool, the code being generated is prom inently shown. Ifyou’re m ore about coding, SAS Studio is likely going to feel m orecom fortable. SAS Enterprise Guide continues to be enhanced to add code-friendly features without taking away the point and click.Flexible Tab LayoutsBoth SAS Studio and SAS Enterprise Guide do a good job letting you m anage your screen real estate. (See “Organizing Your Workspace” in the next section.) SAS Enterprise Guide wins here since, as a desktop application, it has a lot m ore options for viewing m ultiple tabs as well as tearing off tabs to view on another m onitor.Robust ImportingSAS Enterprise Guide has the edge here as there are m any ways to im port m any types of input data via tools and tasks. SAS Studio has a basic im port tool that will continue to be im proved.Microsoft Office IntegrationThis one’s straightforward – SAS Enterprise Guide is a Microsoft Windowsdesktop application, so it can use native Windows protocols to talk toMicrosoft Excel, Microsoft PowerPoint, Microsoft Word, and so on. SAS Studi ois lim ited to talking through the SAS server to com m unicate with MicrosoftOffice applications and relies on a web browser to upload and download files.SAS Enterprise Guide has by far the better integration.Background SubmitSAS Studio 3.8 (and eventually SAS Studio 5.x) allows you to right-click on a .sas file and execute it outside of the interface. This m eans you can disconnect without waiting for it to finish. The next tim e you log on, you can see the results and log that are placed in the sam e location as the .sas file. Really handy for long running jobs or just to be able to do several things at once. This functionality is not yet available in SAS Enterprise Guide. Sharable TasksBoth SAS Studio and SAS Enterprise Guide allow you to create your own tasks – that is, front ends to your code that you can give to others allowing them to run your code and, optionally, prom pt them for changes to the code. Creating SAS Enterprise Guide tasks requires you to use .NET program m ing (ex. C# or ). This is not typically in the skill set of m ost SAS users. As a com piled language, you have full control, but there is a steeper learning curve. SAS Studio tasks, on the other hand, are designed for SAS program m ersand rely on sim ple XML. With sam ple tasks to copy as a starting point and m any papers and docum entation to support task writing, it’s easy to get started. An im portant point to know is that SAS Enterprise Guide currently supports the use of built-in SAS Studio tasks; it does not yet support user-defined SAS Studio tasks, but it will in the future.GitGit is a source m anagem ent system used to track and m anage changes toassets like .sas files, which is especially useful when m ore than oneprogram m er is working on the sam e set of assets. Many organizations are standardizing their work using Git. SAS Enterprise Guide has basic Gitfunctionality, such as com m itting changes, viewing history, and com paringversions. Other operations such as cloning a repository, branching, pushingchanges, and pulling changes m ust be done outside of the SAS EnterpriseGuide application. SAS Studio 3.8 (and eventually SAS Studio 5.x) has m orerobust Git functionality, allowing you to work with a repositoryapplication. You can clone a repository, push and pull changes, m anage branching andm erging – all from the SAS Studio interface. Both applications support working with rem ote repositories (such as GitHub and Atlassian Bitbucket), but SAS Studio currently providesm ore seam less integration. This functionality will eventually be added to SAS Enterprise Guide.WHAT ARE THE SPECIFIC FEATURE DIFFERENCES?The following table lists what features exist in each application called out by release. To explain the release headings, you’ll need to know that at the tim e of this paper, there are two releases of SAS Studio available on SAS Viya (SAS Studio 4.4 and SAS Studio 5.1), one release available on 9.4 (SAS Studio 3.8), and one release of SAS Studio under developm ent that is planned to run on SAS 9.4 and SAS Viya (SAS Studio 5.x). The plan is for SAS Studio 5.x to replace the SAS Studio 4.x series, so SAS Studio 4.x is not shown in the table.SAS Enterprise Guide 8.1 is due to be released m id-2019 on SAS 9.4. This release brings m any SAS Studio features (like the ability to code without a process flow or project) to SAS Enterprise Guide. Ultim ately, like SAS Studio, SAS Enterprise Guide plans a release that will run on SAS 9.4 and on SAS Viya, which is not shown on this chart.It’s im portant to note that this inform ation will change, so for the m ost up-to-date version, visit /software/products/sas-studio/faq/SASStudio_vsEG.htm.Table 1: Feature Comparison Table8.1 3.8 5.1 5.x Standard SAS program m ing ✔✔✔✔✔Interactive SAS program m ing 1✔✔✔XML editor ✔✔✔DATA Step debugger ✔✔✔Append log ✔✔✔Background subm it ✔✔Show list of open item s ✔✔✔Com m and-line interface ✔Console ✔✔Git integration ✔✔✔✔Keyboard m anagem ent ✔✔✔Schedule ✔✔✔Stored process support ✔✔Inform ation Map support ✔✔OLAP cube support ✔✔Tab layout ✔✔✔✔Quick filter ✔✔✔✔Expression builder ✔✔✔✔Freeze colum ns ✔✔Query Builder ✔✔✔✔ Joins ✔✔✔✔ Calculated colum ns ✔✔✔ Where ✔✔✔✔ Having ✔✔✔Im port ✔✔✔✔Max # of rows per page 50010001000SAS defined ✔✔✔✔✔Custom, user-defined .NET XML XML XML Hide/show ✔✔Organize tasks ✔✔Type-ahead filter ✔✔✔✔Custom Code Insertion Points ✔✔SnippetsSAS defined ✔✔✔Custom, user-defined ✔✔✔✔✔Abbreviations ✔✔✔✔Hide/show ✔✔Organize snippets ✔✔Type-ahead filter ✔✔Project 3✔✔Process flows ✔✔✔✔Sub flows ✔✔Reference files ✔✔Em bed files ✔✔✔✔Conditional processing ✔✔SAS Drive on SAS Viya ✔✔Im port SAS Enterprise Guideprojects ✔✔Use SAS Studio Tasks in SASEnterprise Guide2✔✔1 SAS Studio - improved interactive experience in 5.12 You cannot yet use custom SAS Studio tasks in SAS Enterprise Guide3 SAS Enterprise Guide 8.1 – projects are optionalWHAT CAN I DO TO BE A MORE EFFICIENT USER?In this section, you’ll get som e tips and, m ore im portantly, links to m ore info. While links are kind of annoying in a paper, we've included them because m any of these assets are living docum ents that guarantee you'll have access to the latest and greatest inform ation. And there’s lots m ore out there than what’s referenced here – just google “sas enterprise guide tips and tricks” or “sas studio tips and tricks.”ORGANIZING YOUR WORKSPACEThis could be subtitled “m axim izing your real estate.” You m ight be lim ited to one m onitor or you m ight have six, but inevitably you want to see a lot m ore than one thing at a tim e. Som ething that’s new in SAS Enterprise Guide 8.1 is the ability to have m ultiple tabs of inform ation open and drag them around as you like. This is som ething you’ve been able to do in SAS Studio, though m any people don’t realize this.https://com m /t5/SAS-Com m unities-Library/SAS-Studio-tip-How-to-Layout-Tabs/ta-p/476111is an excellent overview. SAS Studio is lim ited to dragging things around within your browser window. If you have m ultiple m onitors, you can open m ultiple browser windows each with its own SAS Studio session, or you can drag a browser window across m onitors, then arrange your tabs. But since SAS Enterprise Guide is a desktop app, thereare a lot m ore options. SAS Enterprise Guide 8.1 allows dragging tabs all around the app or you can just peel off a tab and drag it to a com pletely different m onitor.https://com m /t5/SAS-Com m unities-Library/The-Future-of-SAS-Enterprise-Guide-and-SAS-Studio-a-SAS-Global/ta-p/457437has several exam ples.MAKING THE MOST OF THE EDITORIt’s fastest to just point you to two excellent articles: for SAS Enterprise Guide, see - https:///content/sasdum m y/2017/07/03/sas-program-editor-tricks/and for SAS Studio, see - https://com m /t5/SAS-Com m unities-Library/Tips-for-Program m ing-Efficiently-in-SAS-Studio/ta-p/239367. There are differences in the editors currently but that will change in the future so that you won’t have to rem em ber. AVOIDING TYPING AS MUCH AS POSSIBLEBoth SAS Enterprise Guide and SAS Studio offer keyboard shortcuts. This is a com bination of keys that, when pressed, provide functionality you com m only use – like upcasing a string of text or wrapping com m ent sym bols around a piece of code. There are a lot m ore than you m ight realize since keyboard shortcuts are an integral piece of accessibilityrequirem ents, m aking the interface usable to folks with disabilities and folks who just don’t want to m ove their hands from their keyboard. Go to this link to learn about a few in SAS Enterprise Guide - https:///content/sasdum m y/2013/10/29/five-keyboard-shortcuts/. For SAS Studio, just see the keyboard shortcuts section of the user’s guide. You can insert code snippets into the editor with a few keystrokes, which is handy for things you use a lot. In SAS Enterprise Guide, create an abbreviation. (See the Program Editor tricks blog above.) In SAS Studio, use a keyboard shortcut to insert a saved snippet.SAS Studio also offers the ability to drag table and variable nam es into code from the libraries pane. (See the tips for program m ing efficiently in SAS Studio article above.)WHA T ELSE CA N I USE?Since we work on SAS Enterprise Guide and SAS Studio, we’d love to say that those are your best options for writing SAS code. And while it’s true that they offer lots of goodness specific to SAS, such as interactive syntax help, autocom plete, and easy access to SAS libraries and data sets, it’s also true that there are m any excellent IDE’s and editors out there. Som e can be integrated with SAS easily and som e you can just use as a separate editor and then subm it the finished code to SAS non-interactively or in one of the SAS interfaces.Jupyter is an exam ple of a popular interface that has good integration options for SAS. Jupyter is a freely available notebook-style interface – Jupyter Notebook, JupyterLab, and JupyterHub are exam ples. It supports m any different languages, so to use it to write and subm it SAS code, you go get the SAS kernel from GitHub.https:///content/sasdum m y/2016/04/24/how-to-run-sas-program s-in-jupyter-notebook/is a good overview with links to all you need. You can even use JupyterLab in SAS® University Edition. It’s worth noting that since the notebook style ism ore and m ore in dem and, a notebook perspective is planned for SAS Studio.If you already have a favorite IDE or editor, check to see if there are supporting m aterials to m ake it m ore friendly to use with SAS code. You’ll still likely need to subm it the code outside of the editor but you’ll still get to benefit from the editor features you’ve com e to love. Notepad++is popular andhttps:///content/sasdum m y/2017/08/25/npp-with-sas/gives you som e tips for using it with SAS. IDM UltraEdit is another one with features that are friendly to SAS.SAS Enterprise Guide has an Open in Windows Default option that m akes it easy to edit program s in one of these (or other) editors and then subm it it from within SAS Enterprise Guide. If there are things about your editor that you think should be included in the SAS interfaces, let us know!CONCLUSIONWe’re looking forward to the tim e when som eone com es across this paper in the future and wonders, “why would anyone have to think about which SAS interface they used?” Maybe at that point we just think about SAS code and it appears on the virtual screen in front of our face. OK, m aybe that’s a little too far in the future. But you really shouldn’t have to think about this type of stuff, you already have enough work to do. So check out the links to see what the progress is and be sure to tell us what you want via the SAS Enterprise Guide or SAS Studio com m unities and SASware ballot on com m . ACKNOWLEDGMENTSThank you to Chelsea Mayse for designing the im ages, Chris Hem edinger for the blogs, and Marie Dexter and Jennifer Tam burro for the FAQs on . The biggest thanks go to the enthusiastic and vocal users of SAS Enterprise Guide and SAS Studio who have been driving the direction of all of this for m any years.CONTACT INFORMATIONYour com m ents and questions are valued and encouraged. Contact the authors at: Am y PetersSAS Institute, Inc.+1 919-531-7325Am****************Sam antha DuPontSAS Institute, Inc.+1 919-531-0824Sam********************SAS and all other SAS Institute Inc. product or service nam es are registered tradem arks or tradem arks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration.Other brand and product nam es are tradem arks of their respective com panies.。

pinterest成功的背后学习

可观察性
只要通过facebook或twitter注册，就能够让他人知道使用Pinterest的情况。
感知风险没有任何风险。
第3页/共9页
第三页，编辑于星期六：十四点五十九分。
接收者特质
79%女性平均年龄为40.1 妈妈们准新娘心灵手巧的人烹饪爱好者
第4页/共9页
相关数据来源：DoubleClickAdPlanner 第四页，编辑于星期六：十四点五十九分。
接收者意识到的创新需求
归属需求
相互交换意见给朋友送新奇的礼物分享生活中的乐事
第5页/共9页
第五页，编辑于星期六：十四点五十九分。
启示
发掘未被充分挖掘的细分市场。类似成功案例：PATH
简单的体验过程。类似成功案例：韩寒团队《One·一个》
第6页/共9页
第六页，编辑于星期六：十四点五十九分。
@曹将PPTao
19 likes 6 comments 18 repins
成功的背后
第7页/共9页
第七页，编辑于星期六：十四点五十九分。
感谢您的观看！第9页/共9页第九页来自编辑于星期六：十四点五十九分。
为什么
如此火爆？
270 likes 6 comments
31 repins
第2页/共9页
第二页，编辑于星期六：十四点五十九分。
创新特质
相对优势
瀑布流,用户可以不用翻页就往下观看。界面简单,只是图片，仅仅图片。
兼容性
只要安装好插件,就可以将任何网站的图pin过来。
复杂性
操作简单,就"like""repin""comment"三个功能。 “关注喜欢东西，无社交压力。”

shodan手册

Shodan1.介绍Shodan是一个搜索互联网连接设备的搜索引擎，不同于谷歌、必应、百度这些搜索引擎。

用户可以在Shodan上使用Shodan搜索语法查找连接到互联网的摄像头、路由器、服务器等设备信息。

在渗透测试中，是个很不错的神器。

Shodan的搜索流程：1.1.关于数据1.1.1.BannerShodan采集的基本数据单位是banner。

banner是描述设备所运行服务的标志性文本信息。

对于Web服务器来说，其将返回标题或者是telnet登陆界面。

Banner的内容因服务类型的不同而相异。

以下这是一个典型的HTTP Banner：CODE: [ ]HTT 12OServer nginx 1.1.19Dat S , 0O 201002GContent Typ tex htm;charset uContent Length 646Connection kee alive上面的banner 显示该设备正在运行一个1.1.19版本的nginx Web 服务器软件下面是西门子S7工控系统协议的一个banner ，这次返回的banner完全不同，提供了大量的详细数据(有关的固件信息、序列号):CODE: [ ]Copyright Original Siemens EquipmentP nam S7_TurbineModule typ C3Unknown1)Boo LoaderModule E3BG0A.0.Basic Firmware.3.3.8Module nam C3Serial number o module D9U083642013Plant identificationBasic Hardware E3BG0A.0.注搜索的是联网设备运行服务的，不是单一的主机信息。

如果单暴露很多服务信息，在指定特定搜索内容的时候，搜索结果只会出现指定的内容，不会显示其他的服务信息。

1.1.2.设备元数据除了获取banner ，Shodan还可以获取相关设备的元数据，例如地理位置、主机名、操作系统等信息。

菠萝派网站特点调查报告

菠萝派网站特点调查报告菠萝派是目前国内互联网最大的两个纯买家分享社区，主要为女性用户提供分享和交流的平台，菠萝派网站在功能都包括了热点推荐、女装、男装、童装、家居、好店、特价、搭配、达人等功能，但是在实现上也有一些细微的差别，其区别和各自的特点表现在以下方面：特点1 页面布局菠萝派使用的是横向接近100%的布局方式，这样做的好处是可以更充分的利用空间，使页面内容显得更加丰富特点2 页面表现形式菠萝派在表现形式上使用的是仿pinterest的“图钉式”分享，即把分享的图片以类似于图钉钉在墙上的方式随意的摆放，给人一种随意，慵懒的感觉，符合女性用户“逛街时”的特性特点3 首页前端技术在前端，菠萝派采用的是页面内容动态加载技术，当用户在拖动滚轮向下浏览的时候，图片会动态加载。

这种做法的好处是，一是页面加载速度快，二是节约了流量，因为不可能每个用户浏览页面时从头滚到尾的，这种方法很值得我们借鉴。

至于技术上的实现，可以有多种方式，不过使用的时候要注意不同浏览器兼容性不同的问题。

特点4 植入广告、用户体验广告植入方面，新浪与腾讯微博都有比较多的僵尸粉和专业做广告的，其中新浪微博尤其多。

页面布局上，新浪微博和腾讯微博均把两项重要的功能：“写微博”和“刷新”功能放在左上和右上角，操作性和易用性较差；进入微博详情时，腾讯微博提供浮层功能按钮，更好的利用了手机屏幕中间2/3的黄金区域操作体验上，腾讯微博提供首次操作简单引导功能，发送微薄成功后能立马可见，新浪微博需要手动刷新才能见，造成易用性的差距；腾讯微博表情沿用QQ表情，而新浪微博表情显得鸡肋，呆板无趣；腾讯微博提供方便的“存草稿”功能，可在“我的资料”查看，新浪微博保存后不知所踪；腾讯微博大厅采用“块状”模块，避免了用户上下翻页之苦；新浪微博初始页面默认为上次推出页面，腾讯微博则为微博首页，更加方便。

总的来说，腾讯微博给用户带来的操作体验要优于新浪微博。

总结菠萝派通过对用户体验、功能特点等方面的分析可以看出，安卓版微博手机客户端中，腾讯微博优于新浪微博；对用户数量而言，受网页版微博平移的影响，新浪微博用户多于腾讯微博，但腾讯微博依靠QQ庞大用户群和腾讯公司的实力发展迅速，值得关注。

double pattern lib中的定义 -回复

double pattern lib中的定义-回复Double Pattern Lib中的定义Double Pattern Lib，也被称为DPL，是一种设计模式库，旨在提供一个通用的模型来解决软件设计和开发中常见的问题。

它由一系列设计模式组成，可以在不同的应用场景中使用。

本文将从不同的角度逐步回答Double Pattern Lib的定义，并探讨其在软件开发中的实际运用。

一、什么是Double Pattern Lib？Double Pattern Lib是一个集成的设计模式库，其中包含各种设计模式的实现和使用示例。

它的目的是帮助开发人员更高效地构建可维护和可扩展的软件系统。

Double Pattern Lib采用了面向对象的设计原则，通过使用不同的设计模式，可以有效地解决软件开发中的各种问题。

二、Double Pattern Lib的核心思想Double Pattern Lib的核心思想是提供一种通用的模型来解决重复出现的设计和开发问题。

它通过抽象出公共的设计模式，使开发人员可以在不同的应用场景中使用相同的解决方案。

这种模块化的设计方式可以提高系统的可重用性和灵活性。

三、Double Pattern Lib的分类Double Pattern Lib可以根据功能和用途分为不同的分类，如创建型、结构型和行为型模式。

创建型模式主要关注对象的创建过程，包括工厂模式、单例模式和原型模式等。

结构型模式主要关注对象之间的关系和组织方式，包括适配器模式、装饰器模式和桥接模式等。

行为型模式主要关注对象之间的相互作用，包括观察者模式、策略模式和命令模式等。

四、Double Pattern Lib的优点使用Double Pattern Lib可以带来一系列的优点。

首先，它提供了可重用的解决方案，使开发人员能够快速构建系统。

其次，它提供了一种标准化的设计模式库，使团队成员之间的沟通更加便捷明确。

此外，Double Pattern Lib还提供了一种面向对象的设计风格，使系统更易于维护和扩展。

shardingjdbc水平分表策略

shardingjdbc水平分表策略ShardingJDBC水平分表策略ShardingJDBC是一款开源的分布式数据库中间件，它提供了水平分表策略来解决数据量过大时的扩展性问题。

水平分表是一种将数据划分到多个表中的策略，通过将数据分散存储在不同的表中，可以提高数据库的查询性能和扩展能力。

下面将详细介绍ShardingJDBC水平分表策略的原理和应用。

1. 水平分表概述水平分表是将一张大表按照某种规则拆分成多个小表的过程。

通常情况下，可以根据数据的某个字段（如用户ID、时间戳等）进行拆分，使得相同字段的数据存储在同一个表中，从而提高查询性能。

水平分表的优势在于能够充分利用多台服务器的计算和存储能力，同时降低单表的数据量，减少数据库的查询和写入负载。

2. ShardingJDBC水平分表策略ShardingJDBC提供了多种水平分表策略，其中常用的有按范围分表和按哈希分表两种。

2.1 按范围分表按范围分表是将数据按照某个字段的取值范围进行划分，例如按照订单创建时间划分。

可以根据时间的不同粒度（年、月、日等）创建不同的表，将不同时间段的订单数据存储在不同的表中。

这样的好处是可以根据业务需求，灵活地查询某个时间段内的订单数据，避免全表扫描。

同时，按范围分表也可以根据数据增长的情况，动态地添加新的表，实现数据库的水平扩展。

2.2 按哈希分表按哈希分表是根据某个字段的哈希值进行分表，例如按照用户ID进行哈希分表。

可以将用户ID取哈希值后的结果与分表数取模，将不同的用户数据分散存储在不同的表中。

这样的好处是可以均匀地分散数据到不同的表中，避免某个表的数据过于庞大，影响查询性能。

同时，按哈希分表也可以根据实际情况，动态地调整分表数，实现数据库的弹性扩展。

3. ShardingJDBC水平分表配置使用ShardingJDBC进行水平分表配置，需要在配置文件中指定分表的规则，例如按照订单ID进行哈希分表的配置如下：```shardingRule:tables:order:actualDataNodes: ds${0..1}.order_${0..15}tableStrategy:inline:shardingColumn: order_idalgorithmExpression: order_${order_id.hashCode()%16} ```在上述配置中，actualDataNodes用来指定实际的数据节点，ds${0..1}.order_${0..15}表示分散在多个数据库实例的多个表中。

pinterrest

pinterrest
Pinterest是一个社交网络，可以帮助你将创意变成实际行动。

它提供了一个平台，让你把想法变成真实的行动，无论是个人项目，你的工作还是家庭聚会，Pinterest都有助于跟踪它们并实现它们。

首先，你可以使用Pinterest来创建自己的板条，在这里你可以收藏网页，图片，视频，音乐，等等，来收藏你喜欢的内容。

每个人可以创建多个板条，比如说健康饮食，旅游，园艺，时尚，美容，等等，存放在其中。

你可以从站上的任何页面或者你朋友的页面收藏，还可以从你认识的其他Pinterest用户共享板条，从而让你有更多的内容可以收藏。

此外，Pinterest还提供了一个交流平台，你可以分享你的想法，发现新奇有趣的东西，欣赏来自他人的图片，收集他们的灵感。

你可以在这里和朋友交流，扩大你的影响力，使用hashtags来分类，找到你想要的内容，还能发现以前看不到的新创意。

最后，Pinterest可以帮助你实现创意。

你可以通过搜索功能，在站上找到你想要的创意，你也可以通过团队协作创建一个破局性创意。

Pinterest也有一套自动化系统，可以帮助你实现你的创意，比如你设计一个新网站，或者你想准备个新聚会，这些都可以帮助你实现。

总之，Pinterest是一个很棒的工具，可以帮助你使想法变成现实，它不仅可以让你分享心得，还提供了一个实现创意的工具。

它的卓越的功能，它的先进的工具，让你可以跟踪，实现，完成你的创意，
让你可以把灵感转化为实际行动。

Stable Diffusion 图生图功能介绍

Resize mode
1. Just resize 2. Crop and resize 3. Resize and fill 4. Just resize (latent upscale)
示例 - 新垣结衣照片
1. 选择原因 2. 生成参数
图生图的局限性
1. 需要对模型、微调模型、提示词、参数等有要求 2. 需要不断尝试总结经验
局部绘制 - 涂鸦蒙版 (Inpaint sketch)
1. 结合局部绘制和绘图 2. 示例 - 新垣结衣照片涂鸦蒙版
局部绘制 - 上传蒙版 (Inpaint upload)
1. 使用其他工具制作蒙版并上传 2. 适合专业用户
感谢！
1. 涂粉色头发 2. Denoising strength 的影响
示例 - 从零画一幅画
1. 画出预期场景 2. Denoising strength 的影响
局部绘制 (Inpaint)
1. 指定特定区域进行修改 2. 适用于换脸、换衣服、换背景等场景
参数介绍
1. Mask blur 2. Mask mode 3. Masked content 4. Inpaint area 5. Only masked padding, pixels
不同的 Denoising strength 效果
1. 随着 Denoising strength 变大，越来越不像原图
使用微调模型
1. VAE、Lora 和 HyperNetwork 效果
绘图 (Sketch)
1. 适合有美术基础的用户 2. 添加启动参数，重启 webui
示例 - 新垣结衣照片涂色
Stable Diffusion 图生图功能介绍

sharding proxy 复合分片算法

sharding proxy 复合分片算法
Sharding Proxy是一个基于MySQL的中间件，用于分片和代理数据库操作。

它支持多种分片算法，包括复合分片算法。

复合分片算法是一种自定义的分片算法，它允许用户根据多个字段的值进行分片。

具体来说，复合分片算法通过将多个字段的值组合成一个唯一的标识符，然后将该标识符映射到指定的分片上。

使用复合分片算法，用户可以按照业务需求将数据分散到不同的分片上，以提高查询性能和数据管理效率。

同时，Sharding Proxy还支持对复合分片算法的自定义扩展，以满足不同业务场景的需求。