How to export and reload the model, and add image training on this basis

[GreenHandLJ]

I want use "ImageFolder" as "DataSet" to create and train a new Model then export it,
when I have some new pictures, I can reload this model and train these new pictures on this model

I use different picture to predict ,but all results are the same.
Where can I find the demo

after directory:
models
shoeclassifier-0010.params
synset.txt

result:
[
class: "大蒜", probability: 1.00000
class: "包菜", probability: 0.0e+00
]

Training:
public final class Training {

private static final int BATCH_SIZE = 32;
private static final int EPOCHS = 10;
public static final int IMAGE_HEIGHT = 416;
public static final int IMAGE_WIDTH = 416;

public static void main(String[] args) throws IOException, TranslateException, MalformedModelException {
    Path modelDir = Models.MODEL_PATH;
    ImageFolder dataset = initDataset(new File("E:\\ai\\ai\\data\\training\\min1").toPath());//size 2
    ImageFolder datasetValiddate = initDataset(new File("E:\\ai\\ai\\data\\validate\\min1").toPath());//size 2

// ImageFolder dataset = initDataset(new File("E:\ai\ai\data\training\min2").toPath());//size 1
// ImageFolder datasetValiddate = initDataset(new File("E:\ai\ai\data\validate\min2").toPath());//size 1
TrainingConfig config = setupTrainingConfig(Loss.softmaxCrossEntropyLoss());
Model model = Model.newInstance(Models.MODEL_NAME);
Block resNet50 =
ResNetV1.builder() // construct the network
.setImageShape(new Shape(3, IMAGE_HEIGHT, IMAGE_WIDTH))
.setNumLayers(50)
.setOutSize(3)
.build();
model.setBlock(resNet50);
//load exits model
if (modelDir.toFile().exists()) {
model.load(modelDir, Models.MODEL_NAME);
}

    Trainer trainer = model.newTrainer(config);
    trainer.setMetrics(new Metrics());
    Shape inputShape = new Shape(1, 3, IMAGE_HEIGHT, IMAGE_WIDTH);
    trainer.initialize(inputShape);

    //train
    EasyTrain.fit(trainer, EPOCHS, dataset, datasetValiddate);


    TrainingResult result = trainer.getTrainingResult();
    model.setProperty("Epoch", String.valueOf(EPOCHS));
    model.setProperty(
            "Accuracy", String.format("%.5f", result.getValidateEvaluation("Accuracy")));
    model.setProperty("Loss", String.format("%.5f", result.getValidateLoss()));

    //save model
    model.save(modelDir, Models.MODEL_NAME);
    Models.saveSynset(modelDir, dataset.getSynset());
}

private static ImageFolder initDataset(Path datasetPath)
        throws IOException, TranslateException {
    ImageFolder dataset =
            ImageFolder.builder()
                    // retrieve the data
                    .setRepositoryPath(datasetPath)
                    .optMaxDepth(10)
                    .addTransform(new Resize(IMAGE_WIDTH, IMAGE_HEIGHT))
                    .addTransform(new ToTensor())
                    // random sampling; don't process the data in order
                    .setSampling(BATCH_SIZE, true)
                    .build();

    dataset.prepare();
    return dataset;
}

private static TrainingConfig setupTrainingConfig(Loss loss) {
    return new DefaultTrainingConfig(loss)
            .addEvaluator(new Accuracy())
            .addTrainingListeners(TrainingListener.Defaults.logging());
}

}

predict:
public class Inference {

public static void main(String[] args) throws ModelException, TranslateException, IOException {
    Image img = ImageFactory.getInstance().fromFile(new File("E:\\ai\\ai\\data\\training\\min1\\包菜\\1包菜.jpg").toPath());

    try (Model model = Models.getModel()) { // empty model instance
        // load the model
        model.load(Models.MODEL_PATH, Models.MODEL_NAME);

        Translator<Image, Classifications> translator =
                ImageClassificationTranslator.builder()
                        .addTransform(new Resize(Models.IMAGE_WIDTH, Models.IMAGE_HEIGHT))
                        .addTransform(new ToTensor())
                        .optApplySoftmax(true)
                        .build();

        try (Predictor<Image, Classifications> predictor = model.newPredictor(translator)) {
            Classifications predictResult = predictor.predict(img);
            System.out.println(predictResult);
        }
    }
}

}

[zachgk]

Most of your code seems right, but I am not quite sure what exactly your question is. However, I notice that your dataset has 2 elements and you need to have much more than that. As a comparison, MNIST which is a pretty basic image dataset has 60,000 training images and 10,000 test images.

To retrain a model, you can load it with model.load(...) and then run the usual training process. Make sure to skip the trainer.initialize(...) which puts random values into the model.