search

Hans1984 / TAF

This is the code for siggrapha paper "An Implicit Neural Representation for the Image Stack: Depth, All in Focus, and High Dynamic Range"
MIT License
8 stars 1 forks source link

Training issue #2

Open Janebanana opened 6 months ago

Janebanana commented 6 months ago

Thank you for sharing the interesting work!

I tried to train the model on the datasets provided. Were the different scenes trained with the same setting? For example, I trained the model on real_all/scene_4 without changing anything and could not get reasonable results. Also, how to train the model on the real_dfd data? Should I remove the tone mapping network and how? Looking forward to hearing from you.

Hans1984 commented 6 months ago

Hi, thanks for your attention. For 1, could you provide more information. For 2, yes, in general, just remove the tmo part line 178 to 193, and change the loss function.

Janebanana commented 6 months ago

Thank you for your reply!

  1. I retrained the model and the problem is resolved. I might have made some changes without attention.
  2. Since the current blur image before tone mapping is HDR, do I also need to remove torch.exp(all_in_focus) to deal with that?