I hope this message finds you well. I've been studying your code. In the original code at dynamic-network-architectures/dynamic_network_architectures/building_blocks/unet_decoder.py on line 53, I noticed that both the kernel_size and stride parameters of the transposed convolution are set using the stride value of the corresponding stage in the encoder. This approach seems a bit different from the common practice, where we typically set the kernel_size of the transposed convolution to match the kernel_size of the corresponding stage in the encoder. I was wondering if this design choice was intentional or perhaps an oversight?
Thank you for your time and consideration. I look forward to your response.
I hope this message finds you well. I've been studying your code. In the original code at dynamic-network-architectures/dynamic_network_architectures/building_blocks/unet_decoder.py on line 53, I noticed that both the kernel_size and stride parameters of the transposed convolution are set using the stride value of the corresponding stage in the encoder. This approach seems a bit different from the common practice, where we typically set the kernel_size of the transposed convolution to match the kernel_size of the corresponding stage in the encoder. I was wondering if this design choice was intentional or perhaps an oversight?
Thank you for your time and consideration. I look forward to your response.