search

nex-mpi / nex-code

Code release for NeX: Real-time View Synthesis with Neural Basis Expansion
MIT License
594 stars 73 forks source link

Enquiry regarding the calculation of evaluation results #19

Open derrick-xwp opened 3 years ago

derrick-xwp commented 3 years ago

Dear author,

Thanks for providing us such an excellent work.

I found that the on scene seasoning, the PSNR result of evaluations on 8 views are 0.3 db higher than the result shown in the paper.

I want to confirm with you the calculation method of your evaluation results.

Do you use 8 views per scene to calculate the average of PSNR?

Thanks

pureexe commented 3 years ago

It's not 8 views per scene. it's one of every 8 images. we sort images by name and then pick the image index 0, 8, 16, 24, ...

The image that we use to test the score on the seasoning scene is located in this directory.

https://vistec-my.sharepoint.com/:f:/r/personal/pakkapon_p_s19_vistec_ac_th/Documents/public/VLL/NeX/shiny_datasets/shiny_eval/seasoning?csf=1&web=1&e=A1wwEF

The code that separates the test set. https://github.com/nex-mpi/nex-code/blob/eeff38c712ac9a665f09d7c2a3fdf48ae83f4693/utils/utils.py#L88