search

Drexubery / ViewCrafter

Official implementation of "ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis"
Apache License 2.0
960 stars 34 forks source link

Evaluation setting in Table 1 #44

Closed hangg7 closed 2 weeks ago

hangg7 commented 2 weeks ago

Hi! Thanks for open-sourcing your code and congrats on this strong project!

I am trying to evaluate and compare our method to ViewCrafter in your Table 1. image In your paper, you have described your setting in Section 4.2, but a few details are missing. image

I understand you guys are going to release the evaluation code after CVPR #32. Could you please at the meantime clarify on these questions?

  1. Which CO3D, RealEstate10K test scenes do you use?
  2. What are the easy and hard set splits? I imagine subsampling with different strides, but can you be more specific?
  3. For each scene, what are the condition views and the target views?
  4. What is the resolution of the image, do you perform any resizing or cropping?

Thanks again for your time, and best of the luck for CVPR!

Best, Hang

hangg7 commented 2 weeks ago

Resolved thru email! Thanks @Drexubery!