概述
Deep Java Library是AWS在2019年推出的深度學(xué)習(xí)Java庫(kù)，目前已經(jīng)支持MXNet帘不、PyTorch、TensorFlow模型的訓(xùn)練和推理。DJL沒(méi)有和固定的深度學(xué)習(xí)框架綁定，因此同一套代碼可以適配不同的深度學(xué)習(xí)框架枝哄。
這里根據(jù)官網(wǎng)給的教程，介紹如果搭建目標(biāo)檢測(cè)的Demo阻荒，實(shí)現(xiàn)的功能包括讀取本地圖片挠锥，加載官方Model Zoo提供的預(yù)訓(xùn)練模型、進(jìn)行模型推理侨赡、輸出目標(biāo)檢測(cè)的結(jié)果圖到本地蓖租。參考資料包括：SSD模型推理的官方教程、Maven依賴配置羊壹、DJL Maven的BOM配置蓖宦、DJL版本依賴項(xiàng)搭配。
工程搭建
新建Maven項(xiàng)目
JDK>=1.8
項(xiàng)目結(jié)構(gòu)
依賴引入
引入djl本身以及djl依賴的其他包油猫。
<!-->設(shè)定maven編譯使用jdk8<-->org.apache.maven.pluginsmaven-compiler-plugin88<!-->以BOM的方式統(tǒng)一管理依賴包的版本<-->ai.djlbom0.9.0pomimportcommons-clicommons-cli1.4com.google.code.gsongson2.8.5<!-->日志依賴包<-->org.apache.logging.log4jlog4j-slf4j-impl2.12.1<!-->使用djl必須引入的依賴包<-->ai.djlapi<!-->使用不同的深度學(xué)習(xí)框架模型引入不同的依賴包, 將mxnet改為pytorch即可更換深度學(xué)習(xí)框架
? ? ? ? Apache MXNet engine implementation<-->ai.djl.mxnetmxnet-engine<!-->使用不同的深度學(xué)習(xí)框架模型引入不同的依賴包,
Apache MXNet native library<-->ai.djl.mxnetmxnet-native-autoruntime<!-->使用不同的深度學(xué)習(xí)框架模型引入不同的依賴包,
A ModelZoo containing models exported from Apache MXNet<-->ai.djl.mxnetmxnet-model-zoo復(fù)制代碼
日志配置文件
這部分沒(méi)有固定要求稠茂，按實(shí)際需要來(lái)配置就可以，這里沿用從網(wǎng)上找的一份簡(jiǎn)單配置文件log4j2.xml
<?xml version="1.0" encoding="UTF-8"?>復(fù)制代碼
業(yè)務(wù)代碼
引自：https://github.com/awslabs/djl/blob/master/examples/src/main/java/ai/djl/examples/inference/ObjectDetection.java
packageorg.town;importai.djl.Application;importai.djl.ModelException;importai.djl.engine.Engine;importai.djl.inference.Predictor;importai.djl.modality.cv.Image;importai.djl.modality.cv.ImageFactory;importai.djl.modality.cv.output.DetectedObjects;importai.djl.repository.zoo.Criteria;importai.djl.repository.zoo.ModelZoo;importai.djl.repository.zoo.ZooModel;importai.djl.training.util.ProgressBar;importai.djl.translate.TranslateException;importjava.io.IOException;importjava.nio.file.Files;importjava.nio.file.Path;importjava.nio.file.Paths;importorg.slf4j.Logger;importorg.slf4j.LoggerFactory;/**
* An example of inference using an object detection model.
*/publicfinalclassObjectDetection{privatestaticfinalLogger logger = LoggerFactory.getLogger(ObjectDetection.class);privateObjectDetection(){}publicstaticvoidmain(String[] args)throwsIOException, ModelException, TranslateException{? ? ? ? DetectedObjects detection = ObjectDetection.predict();? ? ? ? logger.info("{}", detection);? ? }publicstaticDetectedObjectspredict()throwsIOException, ModelException, TranslateException{? ? ? ? Path imageFile = Paths.get("src/test/resources/dog_bike_car.jpg");? ? ? ? Image img = ImageFactory.getInstance().fromFile(imageFile);? ? ? ? String backbone;if("TensorFlow".equals(Engine.getInstance().getEngineName())) {? ? ? ? ? ? backbone ="mobilenet_v2";? ? ? ? }else{? ? ? ? ? ? backbone ="resnet50";? ? ? ? }? ? ? ? Criteria criteria =? ? ? ? ? ? ? ? Criteria.builder()? ? ? ? ? ? ? ? ? ? ? ? .optApplication(Application.CV.OBJECT_DETECTION)? ? ? ? ? ? ? ? ? ? ? ? .setTypes(Image.class, DetectedObjects.class)? ? ? ? ? ? ? ? ? ? ? ? .optFilter("backbone", backbone)? ? ? ? ? ? ? ? ? ? ? ? .optProgress(newProgressBar())? ? ? ? ? ? ? ? ? ? ? ? .build();try(ZooModel model = ModelZoo.loadModel(criteria)) {try(Predictor predictor = model.newPredictor()) {? ? ? ? ? ? ? ? DetectedObjects detection = predictor.predict(img);? ? ? ? ? ? ? ? saveBoundingBoxImage(img, detection);returndetection;? ? ? ? ? ? }? ? ? ? }? ? }privatestaticvoidsaveBoundingBoxImage(Image img, DetectedObjects detection)throwsIOException{? ? ? ? Path outputDir = Paths.get("build/output");? ? ? ? Files.createDirectories(outputDir);// Make image copy with alpha channel because original image was jpgImage newImage = img.duplicate(Image.Type.TYPE_INT_ARGB);? ? ? ? newImage.drawBoundingBoxes(detection);? ? ? ? Path imagePath = outputDir.resolve("detected-dog_bike_car.png");// OpenJDK can't save jpg with alpha channelnewImage.save(Files.newOutputStream(imagePath),"png");? ? ? ? logger.info("Detected objects image has been saved in: {}", imagePath);? ? }}復(fù)制代碼
輸出結(jié)果
問(wèn)題記錄
問(wèn)題1：JDK版本過(guò)低導(dǎo)致Static interface method calls are not supported at language level '7'
解決辦法：pom配置文件中顯式指定使用jdk8進(jìn)行編譯情妖。
問(wèn)題2：Exception in thread "main" ai.djl.repository.zoo.ModelNotFoundException: No matching model with specified Input/Output type found.
解決辦法：pom配置文件中沒(méi)有引入ai.djl.mxnet:mxnet-model-zoo依賴包導(dǎo)致的睬关，引入依賴即可。
問(wèn)題3：切換為pytorch引擎后出錯(cuò)毡证。
切換方式：
ai.djl.pytorchpytorch-engineai.djl.pytorchpytorch-native-autoruntimeai.djl.pytorchpytorch-model-zoo復(fù)制代碼
報(bào)錯(cuò)信息：
[2021-01-01 23:49:12.420] [WARN] - [main] ai.djl.engine.Engine - Failed to load engine from: ai.djl.pytorch.engine.PtEngineProviderai.djl.engine.EngineException: Failed to load PyTorch native libraryat ai.djl.pytorch.engine.PtEngine.newInstance(PtEngine.java:56) ~[pytorch-engine-0.9.0.jar:?]at ai.djl.pytorch.engine.PtEngineProvider.getEngine(PtEngineProvider.java:27) ~[pytorch-engine-0.9.0.jar:?]at ai.djl.engine.Engine.initEngine(Engine.java:59) [api-0.9.0.jar:?]at ai.djl.engine.Engine.(Engine.java:49) [api-0.9.0.jar:?]at org.town.ObjectDetection.predict(ObjectDetection.java:45) [classes/:?]at org.town.ObjectDetection.main(ObjectDetection.java:36) [classes/:?]Caused by: java.lang.UnsatisfiedLinkError: C:\Users\steel\.djl.ai\pytorch\1.7.0-cpu-win-x86_64\asmjit.dll: Can't find dependent librariesat java.lang.ClassLoader$NativeLibrary.load(Native Method) ~[?:1.8.0_251]at java.lang.ClassLoader.loadLibrary0(ClassLoader.java:1934) ~[?:1.8.0_251]at java.lang.ClassLoader.loadLibrary(ClassLoader.java:1817) ~[?:1.8.0_251]at java.lang.Runtime.load0(Runtime.java:809) ~[?:1.8.0_251]at java.lang.System.load(System.java:1086) ~[?:1.8.0_251]at java.util.stream.ForEachOps$ForEachOp$OfRef.accept(ForEachOps.java:184) ~[?:1.8.0_251]at java.util.stream.ReferencePipeline$3$1.accept(ReferencePipeline.java:193) ~[?:1.8.0_251]at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:175) ~[?:1.8.0_251]at java.util.stream.ReferencePipeline$3$1.accept(ReferencePipeline.java:193) ~[?:1.8.0_251]at java.util.Iterator.forEachRemaining(Iterator.java:116) ~[?:1.8.0_251]at java.util.Spliterators$IteratorSpliterator.forEachRemaining(Spliterators.java:1801) ~[?:1.8.0_251]at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:482) ~[?:1.8.0_251]at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:472) ~[?:1.8.0_251]at java.util.stream.ForEachOps$ForEachOp.evaluateSequential(ForEachOps.java:151) ~[?:1.8.0_251]at java.util.stream.ForEachOps$ForEachOp$OfRef.evaluateSequential(ForEachOps.java:174) ~[?:1.8.0_251]at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234) ~[?:1.8.0_251]at java.util.stream.ReferencePipeline.forEach(ReferencePipeline.java:418) ~[?:1.8.0_251]at ai.djl.pytorch.jni.LibUtils.loadWinDependencies(LibUtils.java:119) ~[pytorch-engine-0.9.0.jar:?]at ai.djl.pytorch.jni.LibUtils.loadLibrary(LibUtils.java:75) ~[pytorch-engine-0.9.0.jar:?]at ai.djl.pytorch.engine.PtEngine.newInstance(PtEngine.java:44) ~[pytorch-engine-0.9.0.jar:?]... 5 moreException in thread "main" ai.djl.engine.EngineException: No deep learning engine found.Please refer to https://github.com/awslabs/djl/blob/master/docs/development/troubleshooting.md for more details.at ai.djl.engine.Engine.getInstance(Engine.java:119)at org.town.ObjectDetection.predict(ObjectDetection.java:45)at org.town.ObjectDetection.main(ObjectDetection.java:36)Caused by: ai.djl.engine.EngineException: Failed to load PyTorch native libraryat ai.djl.pytorch.engine.PtEngine.newInstance(PtEngine.java:56)at ai.djl.pytorch.engine.PtEngineProvider.getEngine(PtEngineProvider.java:27)at ai.djl.engine.Engine.initEngine(Engine.java:59)at ai.djl.engine.Engine.(Engine.java:49)... 2 moreCaused by: java.lang.UnsatisfiedLinkError: C:\Users\steel\.djl.ai\pytorch\1.7.0-cpu-win-x86_64\asmjit.dll: Can't find dependent librariesat java.lang.ClassLoader$NativeLibrary.load(Native Method)at java.lang.ClassLoader.loadLibrary0(ClassLoader.java:1934)at java.lang.ClassLoader.loadLibrary(ClassLoader.java:1817)at java.lang.Runtime.load0(Runtime.java:809)at java.lang.System.load(System.java:1086)at java.util.stream.ForEachOps$ForEachOp$OfRef.accept(ForEachOps.java:184)at java.util.stream.ReferencePipeline$3$1.accept(ReferencePipeline.java:193)at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:175)at java.util.stream.ReferencePipeline$3$1.accept(ReferencePipeline.java:193)at java.util.Iterator.forEachRemaining(Iterator.java:116)at java.util.Spliterators$IteratorSpliterator.forEachRemaining(Spliterators.java:1801)at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:482)at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:472)at java.util.stream.ForEachOps$ForEachOp.evaluateSequential(ForEachOps.java:151)at java.util.stream.ForEachOps$ForEachOp$OfRef.evaluateSequential(ForEachOps.java:174)at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)at java.util.stream.ReferencePipeline.forEach(ReferencePipeline.java:418)at ai.djl.pytorch.jni.LibUtils.loadWinDependencies(LibUtils.java:119)at ai.djl.pytorch.jni.LibUtils.loadLibrary(LibUtils.java:75)at ai.djl.pytorch.engine.PtEngine.newInstance(PtEngine.java:44)... 5 more復(fù)制代碼
解決辦法：未找到电爹。
