Skip to content

Conversation

@ryanontheinside
Copy link
Collaborator

This PR adds an optical flow preprocessor for use with VACE. Apart from usage of raft small, it follows the specifications of VACE.

Currently, tensorrt added as optional dependency which will require users to install via uv sync --group tensorrt. This is optional as it currently conflicts with other pipelines like overworld.

@ryanontheinside ryanontheinside force-pushed the ryanontheinside/feat/optical-flow-tensorrt branch from f5daf04 to ee6a7ff Compare January 20, 2026 18:28
Add optical flow computation pipeline using RAFT model from torchvision.
Supports both PyTorch and TensorRT backends for acceleration.

- Add OpticalFlowPipeline with lazy model loading
- Add TensorRT engine wrapper and ONNX export utilities
- Add helper functions to reduce code duplication (apply_raft_transforms, load_raft_model)
- Register pipeline in registry and pipeline_manager

Signed-off-by: RyanOnTheInside <7623207+ryanontheinside@users.noreply.github.com>
@ryanontheinside ryanontheinside force-pushed the ryanontheinside/feat/optical-flow-tensorrt branch from ee6a7ff to 9285bea Compare January 20, 2026 18:31
Signed-off-by: RyanOnTheInside <7623207+ryanontheinside@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants