ROOT CAUSE？when used nn.Transpose3d , the shape of weight of transpose3d was consistency with nn.Transpose3D in torch.

[Yuxiang1990]

Hi,
refer to https://github.com/NVIDIA-AI-IOT/torch2trt/blob/master/torch2trt/converters/native_converters.py?plain=1#L439,
out_channels = int(weight.shape[0]) should be modified as out_channels = int(weight.shape[1]). ?

For transpose3d, weight was defined (in_channels,outchannels​, ...) which was exactly opposite of nn.conv3d.