hpcaitech/Open-Sora · error · NotImplementedError
Upsample during project_out is not supported for video
Error message
Upsample during project_out is not supported for video
What it means
When the decoder's project_out upsamples by factor 2, it uses a pixel-shuffle upsampler that is not implemented for 5D video tensors; building a video DCAE with factor 2 project_out raises NotImplementedError. Video configs must use the video-capable project_out setting.
Source
Thrown at opensora/models/dc_ae/models/dc_ae.py:365
build_norm(norm, in_channels),
build_act(act),
]
if factor == 1:
layers.append(
ConvLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
stride=1,
use_bias=True,
norm=None,
act_func=None,
is_video=is_video,
)
)
elif factor == 2:
if is_video:
raise NotImplementedError("Upsample during project_out is not supported for video")
layers.append(
build_upsample_block(
block_type=upsample_block_type, in_channels=in_channels, out_channels=out_channels, shortcut=None
)
)
else:
raise ValueError(f"upsample factor {factor} is not supported for decoder project out")
return OpSequential(layers)
class Encoder(nn.Module):
def __init__(self, cfg: EncoderConfig):
super().__init__()
self.cfg = cfg
num_stages = len(cfg.width_list)
self.num_stages = num_stages
assert len(cfg.depth_list) == num_stages
assert len(cfg.width_list) == num_stagesView on GitHub (pinned to 7ad6a96a13)
Solutions
- Set project_out_factor to a video-supported value (1, with the video conv path)
- Use a video DCAE config shipped with the repo
- Keep is_video=False for image decoding with factor 2
Example fix
# before project_out_factor: 2 # with is_video: true # after project_out_factor: 1
Defensive patterns
Strategy: validation
Validate before calling
if is_video:
assert project_out_factor == 1, 'factor-2 project_out upsample is not implemented for video' Prevention
- Use video-specific configs
- Cross-validate factor fields with is_video at load time
When it happens
Trigger: Constructing a DCAE decoder with is_video=True and project_out_factor: 2; raised from build_decoder_project_out_block during Decoder __init__.
Common situations: Reusing image-checkpoint configs (which commonly use factor 2 project_out) with is_video enabled for video decoding.
Related errors
- ConvPixelUnshuffle downsample is not supported for video
- ConvPixelShuffle upsample is not supported for video
- Downsample during project_in is not supported for video
- local_module {local_module} is not supported
- block_type {block_type} is not supported
AI-assisted analysis of hpcaitech/Open-Sora@7ad6a96a13 (2026-08-28).
Data as JSON: /api/errors/c5468f51bfa7be0f.
Report an issue: GitHub.