search

Tangshitao / MVDiffusion

MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion, NeurIPS 2023 (spotlight)
480 stars 25 forks source link

Question about test_depth_two_stage #23

Open wqdong8 opened 1 year ago

wqdong8 commented 1 year ago

hi, I want to know if there is something wrong with the script for test_depth_two_stage.sh? When I tried it on scannet, I could only get the results between the first two key frames. If i want to obtain the results of all image sequence, are the results of all key frames generated at once, or are they generated once for every two adjacent key frames? Judging from the current code, it may be the latter?

wqdong8 commented 1 year ago

And for custom dataset, how to obtain key frames?

Tangshitao commented 1 year ago

For 2 stage generation, all the key frames are generated first and then interpolated frames are generated subsequently. For custom dataset, you should compute overlap ratios if it's a video or sample views if it's a mesh.