Abnormal Render Result. about nerf-pytorch

[salykova]

@KerryKou Hey, may be the error occurred because of the new version 1.8.1 of pytorch? Can you please try to run the script with pytorch 1.7.0 or 1.4.0?

[salykova]

@KerryKou did you change the parameters in the config file \configs\lego.txt?

[KerryKou]

@KerryKou Hey, may be the error occurred because of the new version 1.8.1 of pytorch? Can you please try to run the script with pytorch 1.7.0 or 1.4.0?

@salykovaa Thank you for your advices! I have tried to use pytorch 1.7.0 and 1.4.0, but the result kept same. This is part of my environment dependencies:

• torch==1.7.0
• torchvision==0.8.0
• imageio==2.9.0
• imageio-ffmpeg ==0.4.3
• matplotlib==3.3.3
• opencv-python==4.5.1.48

It seems that there is no conflict with the requirements.txt. And I also checked my config files. I am sure that they haven't been modified.

[salykova]

@KerryKou Just to be sure: did you put your "fern_test" folder into the "logs" folder? After running the script you should see something like this in the console:

Found ckpts ['./logs/fern_test/200000.tar']
Reloading from ./logs/fern_test/200000.tar

[KerryKou]

@KerryKou Just to be sure: did you put your "fern_test" folder into the "logs" folder? After running the script you should see something like this in the console:

Found ckpts ['./logs/fern_test/200000.tar']
Reloading from ./logs/fern_test/200000.tar

@salykovaa Yes, I am sure I put it correctly. This is what I got from console:

[salykova]

@KerryKou I downloaded the whole repository from this checkpoint and I used the following dependencies:

torch==1.4.0
torchvision==0.5.0
imageio
imageio-ffmpeg
matplotlib
configargparse
tensorboard==1.14.0
tqdm
opencv-python==4.2.0.34

The script works perfectly and the output images are correct. If you use these dependencies, don't forget to install torchsearched manually via cd torchsearchsorted pip install .

If it doesn't help you, I assume the error occurs, because you use 2 GPUs (but I am not sure, just an assumption)

[KerryKou]

@KerryKou I downloaded the whole repository from this checkpoint and I used the following dependencies:

torch==1.4.0
torchvision==0.5.0
imageio
imageio-ffmpeg
matplotlib
configargparse
tensorboard==1.14.0
tqdm
opencv-python==4.2.0.34

The script works perfectly and the output images are correct. If you use these dependencies, don't forget to install torchsearched manually via cd torchsearchsorted pip install .

If it doesn't help you, I assume the error occurs, because you use 2 GPUs (but I am not sure, just an assumption)

@salykovaa I tried to clone from the repository you provided. As you say, it works perfectly. I will go on to find out what causes the abnormal results. Thank you for your help!

[salykova]

@KerryKou You are welcome! My assumption: since the model was trained with torch 1.4.0, we also need to test it with 1.4.0, but maybe I understood this wrong. Anyway, it would be nice if you could find what causes the abnormal result.

[yenchenlin]

Hi all, thanks for raising up the issues. I've temporarily rolled back the master branch to this checkpoint, and move all the updates to branch dev.

The error could come from other people's pull requests, I will try to debug on my side when I have time too.