Skip to content

Fix CPU offload regressions and optimize UV unwrapping for dense meshes #356

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

Open
undergroundrap wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix CPU offload regressions and optimize UV unwrapping for dense meshes #356

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 7 additions & 0 deletions hy3dgen/shapegen/pipelines.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -248,6 +248,13 @@ def __init__(
self.conditioner = conditioner
self.image_processor = image_processor
self.kwargs = kwargs
self.components = {
'vae': vae,
'model': model,
'scheduler': scheduler,
'conditioner': conditioner,
'image_processor': image_processor
}
self.to(device, dtype)

def compile(self):
Expand Down
14 changes: 8 additions & 6 deletions hy3dgen/shapegen/schedulers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -299,10 +299,11 @@ def step(
self._init_step_index(timestep)

# Upcast to avoid precision issues when computing prev_sample
sample = sample.to(torch.float32)
device = model_output.device
sample = sample.to(device=device, dtype=torch.float32)

sigma = self.sigmas[self.step_index]
sigma_next = self.sigmas[self.step_index + 1]
sigma = self.sigmas[self.step_index].to(device=device, dtype=torch.float32)
sigma_next = self.sigmas[self.step_index + 1].to(device=device, dtype=torch.float32)

prev_sample = sample + (sigma_next - sigma) * model_output

Expand Down Expand Up @@ -457,10 +458,11 @@ def step(
if self.step_index is None:
self._init_step_index(timestep)

sample = sample.to(torch.float32)
device = model_output.device
sample = sample.to(device=device, dtype=torch.float32)

sigma = self.sigmas_[self.step_index]
sigma_next = self.sigmas_[self.step_index + 1]
sigma = self.sigmas_[self.step_index].to(device=device, dtype=torch.float32)
sigma_next = self.sigmas_[self.step_index + 1].to(device=device, dtype=torch.float32)

prev_sample = sample + (sigma_next - sigma) * model_output
prev_sample = prev_sample.to(model_output.dtype)
Expand Down
4 changes: 4 additions & 0 deletions hy3dgen/texgen/utils/uv_warp_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,10 @@ def mesh_uv_wrap(mesh):
if isinstance(mesh, trimesh.Scene):
mesh = mesh.dump(concatenate=True)

# Optimization: Decimate extremely dense meshes to prevent xatlas hang
if len(mesh.faces) > 100000:
mesh = mesh.simplify_quadric_decimation(face_count=80000)

if len(mesh.faces) > 500000000:
raise ValueError("The mesh has more than 500,000,000 faces, which is not supported.")

Expand Down
1 change: 1 addition & 0 deletions requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@ trimesh
pymeshlab
pygltflib
xatlas
fast-simplification
#kornia
#facexlib

Expand Down