search

ZHU-Zhiyu / PZRes-Net

Source Codes of the paper " Hyperspectral Image Super-resolution via Deep Progressive Zero-centric Residual Learning"
MIT License
52 stars 9 forks source link

Question about the implementation of loss function #2

Open haydenhs opened 3 years ago

haydenhs commented 3 years ago

Hello Zhiyu, thanks for sharing your code! I read your paper and did not figure out how the loss function is implemented in code main.py. In TIP version, the loss function for SpeNet is L =|| \mathcal{Z}(X) - \mathcal{Z}(\hat{X})|| +||X-\tilde{X}||. However, in main.py, code for loss function is loss = MaeLoss(fout, hsi_g) + MaeLoss(refined, hsi_g) If I understand correctly, hsi_g refers to X and refined is \tilde{X}. So the second part is same as the loss function. However, I could not figure out how the ZM-norm is implemented. I could not found the 'minus mean value' operation in ZM-norm. Is this the correct way to train the net? Could this be a typo?

ZHU-Zhiyu commented 3 years ago

Thanks for your interests. The both versions are utilized in our implementation. We will also provide the code from TIP version soonly.

haydenhs commented 3 years ago

Thanks for your reply! Is the former one a conference version of PZRes-Net? Could you please provide a link to view/download the paper?