Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Test view #17

Closed
dendenxu opened this issue Jan 13, 2024 · 3 comments
Closed

Test view #17

dendenxu opened this issue Jan 13, 2024 · 3 comments

Comments

@dendenxu
Copy link
Member

I run evc -t test -c configs/projects/realtime4dv/rendering/4k4d_sport1.yaml,configs/specs/eval.yaml in NHR dataset and find that the code test 0th, 25th, and 50th input views (200 frames). Is this consistent with the experiment in the paper? In addtion, the code(basketball) test 0th, 26th, and 52th input views(194 frames).

Originally posted by @GuaGod in zju3dv/EasyVolcap#17

@dendenxu
Copy link
Member Author

The provided models are trained on all input views. For qualitative comparison images, you can email me here for instructions.

@GuaGod
Copy link

GuaGod commented Jan 13, 2024

If training scenes from all input views, is it equivalent to including Ground Truth in the model?

@dendenxu
Copy link
Member Author

dendenxu commented Jan 13, 2024

The released code and models contains bug fixes and improvements, thus they might slightly differ from the paper. For the released trained models, we do not leave out test views to achieve the best viewing quality for code users, which is different from the setting in the paper.

The detailed training setting and training/test view split is illustrated in the paper, where the training and test views does not overlap.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants