Value size reshape

[zimingzhong]

When i run python launch.py --config custom/threestudio-3dgs/configs/gaussian_splatting.yaml --train --gpu 0 system.prompt_processor.prompt="a delicious hamburger"

https://github.com/threestudio-project/threestudio/blob/d85e1f8dfdcc1a8015fbedc6b9c213653a47c219/threestudio/models/guidance/stable_diffusion_guidance.py#L267-L270

alpha = self.alphas[t] ** 0.5
sigma = (1 - self.alphas[t]) ** 0.5
latents_denoised = (latents_noisy - sigma * noise_pred) / alpha
image_denoised = self.decode_latents(latents_denoised)

get error: latents denoise = (latents noisy - sigma * noise_pred) / alpha *** RuntimeError: The size of tensor a (4) must match the size of tensor b (64) at non-singleton dimension 3
i change it to
alpha = self.alphas[t] ** 0.5
alpha = alpha.view(-1, 1, 1, 1)
sigma = (1 - self.alphas[t]) ** 0.5
sigma = sigma.view(-1, 1, 1, 1)
latents_denoised = (latents_noisy - sigma * noise_pred) / alpha
image_denoised = self.decode_latents(latents_denoised)
then the code can run
is that right?

[DSaurus]

Hi @zimingzhong ,

Thank you for pointing out this bug! I have fixed it in the latest commit.

[zimingzhong]

Hi @zimingzhong ,

Thank you for pointing out this bug! I have fixed it in the latest commit.

Thank you for your quick reply. I provide a pull on threestudio. You can directly add it.