caoang327/fwd_code

CVPR 2022: https://caoang327.github.io/FWD/

fwd_code's Issues

Code Release

Hi @Caoang327,
Thanks for your great job! I'd like to know when will you release the code. Thanks sincerely.

The results by the code do not match the reported results

Hi, I tried to run the pre-trained model provided in the repository, but the result is worse than the results reported in the paper (PSNR: ~14, SSIM: ~0.55, LPIPS: ~0.385). The synthesized images look very similar to the pictures in the paper. So, I don't know what is the problem.

Camera parameters

How were the camera parameters of the 4x downsampled DTU dataset images generated? And why aren't they used during inference? (seems like the original dtu_down4/camera.npy is used)

Bad performance when training on scale_factor 200

During training, I found that the scale_factor has a significant impact on the training results. I noticed that some other projects use a scale_factor of 200 when utilizing DTU, and I think this might represent the real scale. Therefore, I modified this parameter, but the model's performance deteriorated significantly, with the depth loss remaining above 0.8 and not decreasing. However, if I set it to default value 100, the depth loss decreases rapidly, and the model performs much better. I would like to ask if you remember this situation and if you have any suggestions on how to fix this issue.

training on resized image

Thanks for your work. I have a question about training on resized images. I want to train on DTU images of size (192,256), and I noticed that the DTU_Dataset code automatically handles the resizing for both image and intrinsics. However, when I resize the image, the PSNR drops to around 13 after 10000 steps of training. I'm unsure why this is happening, as I don't see any theoretical flaws that could result in this.

This is the result of 7000 steps for resized image

This is the result of 7000 steps for original size image

Do you have any clue why this is happening? Thank you!

about the release of FWD-U pretrained model

Hi @Caoang327 , would you release the FWD-U pretrained model please?
Thanks for your great job!