search

Totoro97 / f2-nerf

Fast neural radiance field training with free camera trajectories
https://totoro97.github.io/projects/f2-nerf/
Apache License 2.0
933 stars 69 forks source link

test psnr is low #26

Open hanshumin001 opened 1 year ago

hanshumin001 commented 1 year ago

hi, thanks for your amazing work. when I trained a model using my custom dataset, the psnr of training dataset is 37, but the psnr of test/validation is 10. I have checked and not found any problems. Can you help me analyze the problem?

jelleopard commented 1 year ago

Hello, I need you help, Can you describe how to generate cams_meta.npy？ When I render a scene with dynamic objects, F2-NeRF filters out the dynamic objects, do you know why this is?

Totoro97 commented 1 year ago

Hello, did you use the "wanjinyou.yaml" conifg? If you use that config, could you try to set the term renderer.use_app_emb to false and can train it again? Thanks!

hanshumin001 commented 1 year ago

sorry, follow your method, but the problem is still existing. @Totoro97

Totoro97 commented 1 year ago

Hi, how is the visual quality of the testing images?

hanshumin001 commented 1 year ago

As my custom dataset is randomly divided into two parts, the training and testing images has the same distribution. The visual quality has no problem. Also, my dataset is uploaded into hugging face, you can download and try. The address is https://huggingface.co/datasets/Shumin001/nerf/resolve/main/suannai.zip . Thanks for your help!