Octocat
The GatherElements of MNN converter dosen't support dynamic shape, maybe you should fix the input shape.
@zhreshold Thank you for your great job, but I agree with it, a python custom operator implementation for yolo2 output is better to debug....
@zhangliliang thank you, it works
> please paste the error message sorry, my fault. I save the input data of onnx decoder(the padding by ignore_id), and replace random input with these data to run export_onnx.py...
> I have no idea why this error happened. Just wondering why you prefer use libtorch rather than onnx. If you would like use triton, I suggest use onnx which...
> Onnx can support negative index. JIT maybe not ? You may use other value instead of -1 in this case. thanks, I'll try.
> this error if there is -1 in hyps_pad_sos_eos ？ I have the same problem, in my case is hyps_pad_sos_eos have the value greater than output_dim thanks, I'll try.
> 目前已经支持该算子 great job, thx for all your team's work!
@ijkguo you mean I should filter those no label image before training? I don't know whether the PASCAL VOC 2007 contains such images?
@ijkguo But I meet this issue when I used pascal voc dataset