The verification of torch.nn.ConvTranspose2d and torch.nn.Upsample

[ExcellentHH]

The verification of torch.nn.ConvTranspose2d and torch.nn.Upsample

[ExcellentHH]

Hey @ExcellentHH thanks for flagging this :) will get on it asap. Is there a particular upsampling mode you need ? (Default is "nearest").

Thank you for your prompt response. In fact, this issue arose when I was trying to validate the upsampling operation in the U-Net structure. As mentioned in the URL, "In transposed convolutions we have weights that we learn through back-propagation. In papers I’ve come across all of these upsampling methods for various cases and also in practice you may change your architecture and try all of them to see which works best for your own problem. I personally prefer transposed convolutions since we have more control over it but you may go for bilinear interpolation or nearest neighbor for simplicity as well."

I understand that there are different ways to implement upsampling operation, and I would be grateful if EZKL could support torch.nn.ConvTranspose2d and as many modes (bilinear, nearest, etc.) as possible for torch.nn.Upsample. I believe that this will greatly expand EZKL's application range, and thanks for the hard work.

Please let me know if there is anything I can do to assist with this feature, and I look forward to hearing back from you soon.

[ExcellentHH]

Hey @ExcellentHH rolling out ConvTranspose2D in #263 . Also added your example above as a unit test :)

Thank you for your hard work on developing the new feature. I am eagerly looking forward to testing it as soon as possible!

Once again, I would like to express my sincere thanks for your response and support. Please let me know if there is anything I can do to assist with testing or providing feedback on the new feature. Furthermore, I hope that the EZKL project will continue to flourish and grow.

[alexander-camuto]

Hey @ExcellentHH thanks for flagging this :) will get on it asap. Is there a particular upsampling mode you need ? (Default is "nearest").

[alexander-camuto]

@ExcellentHH I've kicked off work for ConvTranspose2D / Deconv op :) stay tuned

[alexander-camuto]

Hey @ExcellentHH rolling out ConvTranspose2D in #263 . Also added your example above as a unit test :)

[alexander-camuto]

Hey @ExcellentHH #264 implements nearest neighbours upsampling ! hope that helps

[ExcellentHH]

Hey @ExcellentHH #264 implements nearest neighbours upsampling ! hope that helps

Thanks a lot, I'll update my local version to use the new features. Thank you very much for your hard work again!