Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

debug

Shen-Chenhuida5f42d488df
diff --git a/opensora/models/vae/vae_3d.py b/opensora/models/vae/vae_3d.py
index d4a47ad..4bed5ad 100644
--- a/opensora/models/vae/vae_3d.py
+++ b/opensora/models/vae/vae_3d.py
@@ -681,7 +681,7 @@ class Decoder(nn.Module):
# conv blocks with upsampling
if i > 0:
- if self.temporal_downsample[i]:
+ if self.temporal_downsample[i-1]:
t_stride = 2 if self.temporal_downsample[i - 1] else 1
# SCH: T-Causal Conv 3x3x3, f -> (t_stride * 2 * 2) * f, depth to space t_stride x 2 x 2
self.conv_blocks.insert(
@@ -697,7 +697,7 @@ class Decoder(nn.Module):
prev_filters
),
)
-
+
self.norm1 = nn.GroupNorm(self.num_groups, prev_filters, device=device, dtype=dtype)
diff --git a/scripts/inference-vae.py b/scripts/inference-vae.py
index 63030db..b6a4e52 100644
--- a/scripts/inference-vae.py
+++ b/scripts/inference-vae.py
@@ -9,8 +9,8 @@ from tqdm import tqdm
from opensora.acceleration.parallel_states import get_data_parallel_group
from opensora.datasets import prepare_dataloader, save_sample
-from opensora.models.vae.losses import AdversarialLoss, DiscriminatorLoss, VEALoss
-from opensora.models.vae.vae_3d import LeCamEMA, pad_at_dim
+from opensora.models.vae.losses import AdversarialLoss, DiscriminatorLoss, VEALoss, LeCamEMA
+from opensora.models.vae.vae_3d import pad_at_dim
from opensora.registry import DATASETS, MODELS, build_module
from opensora.utils.config_utils import parse_configs
from opensora.utils.misc import to_torch_dtype