Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

config pixabay

Shen-Chenhui8548b88ca804
diff --git a/configs/vae_magvit_v2/train/16x256x256_pixabay_8_GPU_pipeline.py b/configs/vae_magvit_v2/train/16x128x128_pixabay_8_GPU_pipeline.py
similarity index 92%
rename from configs/vae_magvit_v2/train/16x256x256_pixabay_8_GPU_pipeline.py
rename to configs/vae_magvit_v2/train/16x128x128_pixabay_8_GPU_pipeline.py
index 14bf7ba..dab0c98 100644
--- a/configs/vae_magvit_v2/train/16x256x256_pixabay_8_GPU_pipeline.py
+++ b/configs/vae_magvit_v2/train/16x128x128_pixabay_8_GPU_pipeline.py
@@ -1,6 +1,6 @@
num_frames = 16
frame_interval = 3
-image_size = (256, 256)
+image_size = (128, 128)
use_pipeline = True
# Define dataset
@@ -44,13 +44,13 @@ model = dict(
discriminator = dict(
type="DISCRIMINATOR_3D",
- image_size = (32, 32), # NOTE: here image size is different
+ image_size = (16, 16), # NOTE: here image size is different
num_frames = num_frames,
in_channels = 4,
filters = 128,
use_pretrained=True, # NOTE: set to False only if we want to disable load
# channel_multipliers = (2,4,4,4,4), # (2,4,4,4) for 64x64 resolution
- channel_multipliers= (2,4,4) # since on intermediate layer, 16 x 32 x 32 dimension z
+ channel_multipliers= (2,4,4) # since on intermediate layer dimension ofs z
)