In the transposed convolution source code, the output shape is not the same as Pytorch, for same parameters. In order to solve this, I have moved the padding after self.conv call. This is for a tranposed convolution with kernel size [4, 4] and strides [2, 2]. Can you please recheck your conversion code on other tranposed convolutions?
Yes, TFConvTranspose is problematic, only a few kinds of situation are checked.
I have no plan to fix bug for TFConvTranspose in recently, if your code can make TFConvTranspose work better, welcome PR.
Hi,
In the transposed convolution source code, the output shape is not the same as Pytorch, for same parameters. In order to solve this, I have moved the padding after self.conv call. This is for a tranposed convolution with kernel size [4, 4] and strides [2, 2]. Can you please recheck your conversion code on other tranposed convolutions?
https://github.com/MPolaris/onnx2tflite/blob/main/layers/conv_layers.py#L37