Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

debug

Shen-Chenhui0ad337178ae1
diff --git a/opensora/models/vae/vae_3d_v2.py b/opensora/models/vae/vae_3d_v2.py
index d81af23..44983e0 100644
--- a/opensora/models/vae/vae_3d_v2.py
+++ b/opensora/models/vae/vae_3d_v2.py
@@ -438,6 +438,7 @@ class StyleGANDiscriminatorBlur(nn.Module):
time_scaled = num_frames // scale_factor + 1
else:
time_scaled = num_frames / scale_factor
+
assert self.input_size[0] % scale_factor == 0, f"image width {self.input_size[0]} is not divisible by scale factor {scale_factor}"
assert self.input_size[1] % scale_factor == 0, f"image height {self.input_size[1]} is not divisible by scale factor {scale_factor}"
w_scaled, h_scaled = self.input_size[0] / scale_factor, self.input_size[1] / scale_factor
diff --git a/scripts/train-vae-v2.py b/scripts/train-vae-v2.py
index 3d10db0..8ff7045 100644
--- a/scripts/train-vae-v2.py
+++ b/scripts/train-vae-v2.py
@@ -255,7 +255,10 @@ def main():
# calculate discriminator_time_padding
disc_time_downsample_factor = 2 ** len(cfg.discriminator.channel_multipliers)
- disc_time_padding = disc_time_downsample_factor - cfg.num_frames % disc_time_downsample_factor
+ if cfg.num_frames % disc_time_downsample_factor != 0:
+ disc_time_padding = disc_time_downsample_factor - cfg.num_frames % disc_time_downsample_factor
+ else:
+ disc_time_padding = 0
video_contains_first_frame = cfg.video_contains_first_frame
lecam_ema_real = torch.tensor(0.0)
@@ -340,7 +343,7 @@ def main():
if global_step > cfg.discriminator_start:
# padded videos for GAN
fake_video = pad_at_dim(recon_video, (disc_time_padding, 0), value = 0., dim = 2)
- fake_logits = discriminator(fake_video.contiguous()) # TODO: take out contiguous?
+ fake_logits = discriminator(fake_video.contiguous())
adversarial_loss = adversarial_loss_fn(
fake_logits,
nll_loss,