Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

config

Shen-Chenhui0fdc6e99350e
diff --git a/configs/vae_magvit_v2/train/17x128x128.py b/configs/vae_magvit_v2/train/17x128x128.py
index a5e8c03..e315b3e 100644
--- a/configs/vae_magvit_v2/train/17x128x128.py
+++ b/configs/vae_magvit_v2/train/17x128x128.py
@@ -21,7 +21,7 @@ perceptual_weight = 0.1 # TODO: need to change this to 0.1 !!! according to MAGV
# Define model
model = dict(
- type="VAE_3D_B",
+ type="VAE_MAGVIT_V2",
in_out_channels = 3,
latent_embed_dim = 16,
filters = 128,
diff --git a/opensora/models/vae/vae_3d_v2.py b/opensora/models/vae/vae_3d_v2.py
index d9fa2ff..e72f196 100644
--- a/opensora/models/vae/vae_3d_v2.py
+++ b/opensora/models/vae/vae_3d_v2.py
@@ -349,7 +349,7 @@ class Decoder(nn.Module):
@MODELS.register_module()
-class VAE_3D(nn.Module):
+class VAE_3D_V2(nn.Module):
"""The 3D VAE """
def __init__(
self,
@@ -456,9 +456,9 @@ class VAE_3D(nn.Module):
return dec, posterior
[email protected]_module("VAE_3D_B")
-def VAE_3D_B(from_pretrained=None, **kwargs):
- model = VAE_3D(**kwargs)
[email protected]_module("VAE_MAGVIT_V2")
+def VAE_MAGVIT_V2(from_pretrained=None, **kwargs):
+ model = VAE_3D_V2(**kwargs)
if from_pretrained is not None:
load_checkpoint(model, from_pretrained)
return model
\ No newline at end of file