search

sherwinbahmani / 4dfy

4D-fy: Text-to-4D Generation Using Hybrid Score Distillation Sampling
https://sherwinbahmani.github.io/4dfy/
Apache License 2.0
288 stars 7 forks source link

Optimization procedure in stage 3 #11

Closed ByChelsea closed 4 months ago

ByChelsea commented 4 months ago

Thanks again for sharing your work! According to the paper, all three losses are used in the third stage. However, in the code, it seems that VSD is not used in this stage?

sherwinbahmani commented 4 months ago

Hi, In the low vram config it is not used, in the main config the single_view probability is set to 0.5. So it alternates between multi view (0.5) and single view (0.5). And within single view it alternates between video and vsd with 0.5 probability.

ByChelsea commented 4 months ago

Ah, sorry! I didn't carefully look into the differences between these two settings.