How to provide multi label dataset for object detection?

[thhart]

What is the expected structure of AccessDataset for training object detection models. The PikachuDetection contains a nice example however it is a one class example only. Below is the label extraction example constructing a Record structure. How must this structure constructed if there are multiple classes to be trained? Where is the class number to be specified?

    protected Record get(NDManager manager, long index) throws IOException {
        int idx = Math.toIntExact(index);
        NDList d =
                new NDList(
                        ImageFactory.getInstance()
                                .fromFile(imagePaths.get(idx))
                                .toNDArray(manager, flag));
        NDArray label = manager.create(labels.get(idx));
        NDList l = new NDList(label.reshape(new Shape(1).addAll(label.getShape())));
        return new Record(d, l);
    }

[stu1130]

The label type is NDList, which represents a list of NDArray. You can add several labels to construct a label NDList.