Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

debug

Shen-Chenhuidc63a2b4e5fe
diff --git a/opensora/models/vae/vae_3d_v2.py b/opensora/models/vae/vae_3d_v2.py
index 6204618..649fce4 100644
--- a/opensora/models/vae/vae_3d_v2.py
+++ b/opensora/models/vae/vae_3d_v2.py
@@ -609,8 +609,7 @@ class VAE_3D_V2(nn.Module):
# perceptual loss
- if self.use_vgg:
-
+ if self.perceptual_loss_weight is not None and self.perceptual_loss_weight > 0:
frame_indices = torch.randn((batch, frames)).topk(1, dim = -1).indices
input_vgg_input = pick_video_frame(video, frame_indices)
@@ -626,8 +625,8 @@ class VAE_3D_V2(nn.Module):
input_vgg_feats = self.vgg(input_vgg_input)
recon_vgg_feats = self.vgg(recon_vgg_input)
-
perceptual_loss = F.mse_loss(input_vgg_feats, recon_vgg_feats)
+
else:
perceptual_loss = self.zero