Haochen

maybe not only that; there's also the possibility that overall quality could improve as well. Will update under the issue when we introduce more updates in the next few weeks.

Thanks for the encouragement! The issue of front face hallucination (Janus) remains unresolved in many cases, and we are exploring some solutions less reliant on the language model than view-dependent...

Hi Thanks a lot for the encouragement. The vis video looks jittery cuz I forgot to add the final vis function with sub-pixel rendering. Please hold off for a while...

I have added the subpixel rendering script. The jittery video frames should be ok now.

The underlying 3D representation (NeRF) is constraining the system to provide view consistency. It is true that at each iteration, the guidance provided by the 2D diffusion is pretty random....

We will. This uses < 9GB of GPU memory, and should be ok with regular colab.

wow thank you!

@LucasSilvaFerreira no the colab is not using the subpix rendering. we will update the colab.

the converted weight is already provided here https://github.com/w-hc/torch_audioset/releases/download/v0.1/yamnet.pth sry about the lack of clarity.