cvg/nerf-on-the-go

Questions about the evaluate of datasets

Sugar55888 opened this issue · 4 comments

Sugar55888 commented

Hi, thank you for your excellent work! Regarding the evaluation of the nerf-on-the-go dataset, I have some doubts. Because some of the test images contain occlusions, and the in the wild method focuses on removing the distractors and rendering clean images. Will this lead to a decrease in the PSNR value calculated between the rendered image and the test true value image? I don't know if my idea is right. Could you provide some suggestions?

rwn17 commented

Thanks for your interest in our work! Could you please list the index for the test images with occlusion? You are right the test set should be distractor-free.

Sugar55888 commented

@rwn17 Hi, in some highly occluded scenes, such as train_station, almost every picture has distractors. It's hard to select some images for test.
image

rwn17 commented

Yes you are correct, so we do not include this scene in the quantitative evaluation. Please refer to Table 1 for the scenes we use for quantitative assessment.

Sugar55888 commented

@rwn17 Thanks for your answer! I have seen it!

  • 👍1