Question

因此，我在迁移学习示例中工作。我的任务是拍摄VGG-16，使用传递学习对其进行修改和训练，以便仅预测2个标签（“猫”和“狗”）。这是我的代码，它的格式不正确，所以请不要被“混乱”分散注意力：

    static DataSetIterator trainIter;

    public static void main(String[] args) throws Exception {
        SpringApplication.run(AiProjectDl4jApplication.class, args);

        int seed = 12345;
        int numClasses = 2;


        ZooModel zooModel = VGG16.builder().build();
        ComputationGraph pretrainedNet = (ComputationGraph) zooModel.initPretrained(PretrainedType.IMAGENET);

        FineTuneConfiguration fineTuneConf = new FineTuneConfiguration.Builder()
                .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
                .updater(new Nesterovs(5e-5))
                .seed(seed)
                .build();   


        ComputationGraph vgg16Transfer = new TransferLearning.GraphBuilder(pretrainedNet)
                .fineTuneConfiguration(fineTuneConf)
                .setFeatureExtractor("fc2")
                .removeVertexKeepConnections("predictions")
                .addLayer("predictions",
                        new OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD)
                                .nIn(4096).nOut(numClasses)
                                .weightInit(WeightInit.XAVIER)
                                .activation(Activation.SOFTMAX).build(), "fc2")
                .build();

        TransferLearningHelper transferLearningHelper =
                new TransferLearningHelper(vgg16Transfer, "fc2");

        getImages();
        DataSetPreProcessor preProcessor = new VGG16ImagePreProcessor() ;
        trainIter.setPreProcessor(preProcessor);

        vgg16Transfer.setListeners(new ScoreIterationListener(5));
        log.info("Training starting...");

        for (int i = 0; i < 5; i++) {
            while (trainIter.hasNext()) {
                DataSet currentFeaturized = trainIter.next();
                vgg16Transfer.fit(currentFeaturized);
            }
        }
        log.info("If no results shown -> Summary: Failed");   
    }


    static void getImages() throws IOException {
        Random rng = new Random();
        File parentDir = new File("Path To DataSet"); 

        ParentPathLabelGenerator labelMaker = new ParentPathLabelGenerator();

        ImageRecordReader recordReader = new ImageRecordReader(224,224,3,labelMaker);
        recordReader.initialize(new FileSplit(new File(String.valueOf(parentDir))));

        trainIter = new RecordReaderDataSetIterator(recordReader,150,1,2);

    }
}

问题出在我尝试训练修改后的模型的地方。我只得到第一次迭代的结果（已经很高，为0.9171），然后得到此错误：

# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x00007ffb426a3b29, pid=11708, tid=0x0000000000001a78
#
# JRE version: Java(TM) SE Runtime Environment (8.0_251-b08) (build 1.8.0_251-b08)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (25.251-b08 mixed mode windows-amd64 compressed oops)
# Problematic frame:
# C  [KERNELBASE.dll+0x43b29]
#
# Failed to write core dump. Minidumps are not enabled by default on client versions of Windows

几个月前，我已经尝试过建立和训练自己的网络，但遇到了同样的问题，没人能真正帮助我。这次我希望你能。您有什么可能的原因吗？

培训期间，JRE在KERNELBASE.dll + 0x43b29检测到错误

0 个答案: