xuxu116

> > In the training, I got this wrong as follows. > > tensor([[nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan,...

先转onnx再转ncnn