Viv
Results
3
comments of
Viv
```python w = vae_2d_ckpt["state_dict"][key_2d] # conv2d weight new_w = torch.zeros(shape_3d, dtype=w.dtype) new_w[:, :, -1, :, :] = w ``` https://github.com/vivym/OmniGen/blob/main/scripts/inflate_conv_for_video_vae.py
@Birdylx It is indeed an odd number of frames. You can refer to the paper https://arxiv.org/abs/2310.05737
Hope to see any update on it!