the model scrfd_500m_bnkps.onnx predictor error

[Aunixtoyou]

Description

model info
scrfd_500m_bnkps.onnx

inputs
name: input.1
tensor: float32[?,3,?,?]

outputs
score_8
name: score_8
tensor: float32[?,1]
score_16
name: score_16
tensor: float32[?,1]
score_32
name: score_32
tensor: float32[?,1]
bbox_8
name: bbox_8
tensor: float32[?,4]
bbox_16
name: bbox_16
tensor: float32[?,4]
bbox_32
name: bbox_32
tensor: float32[?,4]
kps_8
name: kps_8
tensor: float32[?,10]
kps_16
name: kps_16
tensor: float32[?,10]
kps_32
name: kps_32
tensor: float32[?,10]

Expected Behavior

success

Error Message

Caused by: ai.djl.engine.EngineException: split_size can only be 0 if dimension size is 0, but got dimension size of 3200
at ai.djl.pytorch.jni.PyTorchLibrary.torchSplit(Native Method)
at ai.djl.pytorch.jni.JniUtils.split(JniUtils.java:969)
at ai.djl.pytorch.engine.PtNDArray.split(PtNDArray.java:1068)
at ai.djl.ndarray.NDArrayAdapter.split(NDArrayAdapter.java:876)
at ai.djl.ndarray.NDArray.split(NDArray.java:3191)
at ai.djl.translate.StackBatchifier.unbatchify(StackBatchifier.java:118)
at ai.djl.translate.Translator.batchProcessOutput(Translator.java:119)
at ai.djl.inference.Predictor.batchPredict(Predictor.java:194)
... 4 more

How to Reproduce?

@Override
public NDList processInput(TranslatorContext ctx, Image input) throws Exception {
    NDArray img = input.toNDArray(ctx.getNDManager(), Image.Flag.COLOR);
    img = NDImageUtils.resize(img, 640, 640);
    img = img.transpose(2, 0, 1).flip(0); // HWC -> CHW RGB -> BGR
    NDArray mean =
            ctx.getNDManager().create(new float[]{104f, 117f, 123f}, new Shape(3, 1, 1));
    img = img.sub(mean);
    return new NDList(img);
}

Steps to reproduce

What have you tried to solve it?

no idea

Environment Info

Please run the command ./gradlew debugEnv from the root directory of DJL (if necessary, clone DJL first). It will output information about your system, environment, and installation that can help us debug your issue. Paste the output of the command below:

PASTE OUTPUT HERE

[Aunixtoyou]

在 ai.djl.ndarray.NDArray 中

  @Override
    public NDList[] unbatchify(NDList inputs) {
        int numInputKinds = inputs.size();
        if (numInputKinds == 0) {
            return new NDList[0];
        }
        int batchSize = Math.toIntExact(inputs.head().size(0));
        if (batchSize == 0) {
            return new NDList[0];
        }

        NDList[] dataList = new NDList[batchSize];
        for (int i = 0; i < batchSize; i++) {
            dataList[i] = new NDList();
        }

        for (NDArray input : inputs) {
            NDList splitList = input.split(batchSize);
            for (int i = 0; i < batchSize; i++) {
                NDArray array = splitList.get(i).squeeze(0);
                array.setName(input.getName());
                dataList[i].add(array);
            }
        }
        return dataList;
    }

batchSize=12800
这里所有的inpt都使用同一个batchSize 导致后面的input计算出来的split sectionSize 为0 就会抛出以上异常。请问这里为啥是这样子的呢？