You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This "bug" can be reproduced by doing GroupNorm after CausalConv3d. And we accidentally found that this bug can be temporarily solved by adding: h == torch.zeros_like(h)
before each h = self.norm(h)
in opensora/models/ae/videobase/modules/resnet_block.py
Problem:
Encountered RuntimeError: dim ([32]) cannot convert to NDC1HWC0 when trying to reconstruct a video using rec_imvi_vae.py
Command:
python examples/rec_imvi_vae.py --model_path /LanguageBind/Open-Sora-Plan-v1.0.0 --video_path test_video.mp4 --rec_path rec_video.mp4 --fps 24 --resolution 512 --crop_size 512 --num_frames 64 --sample_rate 1 --ae CausalVAEModel_4x8x8
Encountered the same problem when adding
--enable_tiling
Weights:
/LanguageBind/Open-Sora-Plan-v1.0.0/vae
Devices:
NPU 910A 32GB
The text was updated successfully, but these errors were encountered: