Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Support different strength for Stable Diffusion TensorRT Inpainting pipeline #4216

Merged
merged 2 commits into from
Aug 3, 2023
Merged

Support different strength for Stable Diffusion TensorRT Inpainting pipeline #4216

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a7b00f9
Support different strength
jinwonkim93 Jul 23, 2023
ec166eb
run make style
jinwonkim93 Jul 23, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
run make style
  • Loading branch information
@jinwonkim93
jinwonkim93 committed Jul 23, 2023
commit ec166eb54431c921d329ac9f686adae83a8ab425
32 changes: 16 additions & 16 deletions examples/community/stable_diffusion_tensorrt_inpaint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -829,7 +829,7 @@ def __initialize_timesteps(self, num_inference_steps, strength):
init_timestep = int(num_inference_steps * strength) + offset
init_timestep = min(init_timestep, num_inference_steps)
t_start = max(num_inference_steps - init_timestep + offset, 0)
timesteps = self.scheduler.timesteps[t_start*self.scheduler.order:].to(self.torch_device)
timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :].to(self.torch_device)
return timesteps, num_inference_steps - t_start

def __preprocess_images(self, batch_size, images=()):
Expand Down Expand Up @@ -1043,29 +1043,29 @@ def __call__(
latent_height = self.image_height // 8
latent_width = self.image_width // 8



# Pre-process input images
mask, masked_image, init_image = self.__preprocess_images(batch_size,
prepare_mask_and_masked_image(image,
mask_image,
self.image_height,
self.image_width,
return_image=True,
)
)
mask, masked_image, init_image = self.__preprocess_images(
batch_size,
prepare_mask_and_masked_image(
image,
mask_image,
self.image_height,
self.image_width,
return_image=True,
),
)
# print(mask)
mask = torch.nn.functional.interpolate(mask, size=(latent_height, latent_width))
mask = torch.cat([mask] * 2)

# Initialize timesteps
timesteps, t_start = self.__initialize_timesteps(self.denoising_steps, strength)

# at which timestep to set the initial noise (n.b. 50% if strength is 0.5)
latent_timestep = timesteps[:1].repeat(batch_size)
# create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise
is_strength_max = strength == 1.0

# Pre-initialize latents
num_channels_latents = self.vae.config.latent_channels
latents_outputs = self.prepare_latents(
Expand All @@ -1080,9 +1080,9 @@ def __call__(
timestep=latent_timestep,
is_strength_max=is_strength_max,
)

latents = latents_outputs[0]

# VAE encode masked image
masked_latents = self.__encode_image(masked_image)
masked_latents = torch.cat([masked_latents] * 2)
Expand All @@ -1104,4 +1104,4 @@ def __call__(
images = self.__decode_latent(latents)

images = self.numpy_to_pil(images)
return StableDiffusionPipelineOutput(images=images, nsfw_content_detected=None)
return StableDiffusionPipelineOutput(images=images, nsfw_content_detected=None)
Loading