Hi,
There are so many bugs in snpe-onnx-to-dlc scripts, especially for ConvTranspose.
1,‘’infer_output_shape‘’ should be ''infer_output_shapes'' in nn_translations.py, line 251.
2, Weight shape in deconvolution is CNHW, so the bias length should be equal to weight.shape[1] rather than shape[0] . The mistake lies in nn_translations.py, line 213 : '' bias = numpy.zeros(weights.shape[0], dtype=numpy.float32)''
3, Why not support output padding? How can I get the right output_width and ouput_height? Through params directly?
By the way, do you have a plan to support MXNet?
By the way, do you have a plan to support MXNet?