search

Wanggcong / StyleLight

[ECCV 2022] StyleLight: HDR Panorama Generation for Lighting Estimation and Editing
https://style-light.github.io/
MIT License
125 stars 8 forks source link

The evaluation include Angular Error, RMSE, siRMSE and FID. #4

Closed asmile201711360030 closed 1 year ago

asmile201711360030 commented 1 year ago

hi,I want to know where is the code for calculating MSE and RMSE？

Wanggcong commented 1 year ago

Please check our Evaluation for siMSE and RMSE. For FID, please find the evaluation at this link.

asmile201711360030 commented 1 year ago

Thanks,I have a new question. The MSE in the EMLight paper is on the order of 0.01, while the MSE in your paper is on the order of 0.1. Why is it different？

Wanggcong commented 1 year ago

Thanks,I have a new question. The MSE in the EMLight paper is on the order of 0.01, while the MSE in your paper is on the order of 0.1. Why is it different？

Thanks for the question. 1) Since the parameters of three spheres (material) in EMLight are not released, we set our spheres for rendering. 2) We mask the spheres for evaluation while EMLight computes MSE for the whole image.

asmile201711360030 commented 1 year ago

I have a new question.I used your rendering code and rendered it using HDR images, resulting in an RMSE result of 0.01 level. I am learning about rendering. Do you think there is a problem with this approach?

Wanggcong commented 1 year ago

Sorry, without more detailed description, it could be hard to find the main problem. For example, different spheres with different materials would significantly different results.

One way to debug your result is to visualize rendered results with ground-truth HDR images and your generated HDR images.

Wanggcong commented 1 year ago

If there are no other questions, I will close this issue. You may re-open or open a new issue if any.