WaveCNet icon indicating copy to clipboard operation
WaveCNet copied to clipboard
verified publisher icon LiQiufu

No IDWT layer used in main procedure

Open momo1986 opened this issue 4 years ago • 4 comments

Thanks a lot for sharing. Looks like in the paper, there are three steps:

  1. Do DWT operation
  2. Directly forward low-frequency elements, and filter high-frequency elements.
  3. IDWT to reconstruct the signal.

My question is how to combine the IDWT operation with the end-to-end CNNs?

Thanks & Regards! Momo

momo1986 avatar Apr 20 '21 05:04 momo1986

I applied DWT and IDWT in DeepLab or U-Net for image segmentation. https://github.com/LiQiufu/WaveSNet

LiQiufu avatar Apr 20 '21 05:04 LiQiufu

@LiQiufu Li, thanks.

Therefore, for classification task, only dwt module can work fine. Is my understanding wrong?

Also very grateful to your reply.

Regards! Momo

momo1986 avatar Apr 20 '21 14:04 momo1986

IDWT is for data up-sampling, while no up-sampling operation is used in the deep networks for image classification. So, I don't think it is a good idea to do IDWT in classification.

LiQiufu avatar Apr 21 '21 01:04 LiQiufu

@LiQiufu , I am grateful to your nice reply.

Hopes you are all good.

Best Regards! Momo

momo1986 avatar Apr 21 '21 03:04 momo1986