autonomousvision/unimatch
A unified transformer-based model for optical flow, stereo matching, and monocular depth estimation achieving state-of-the-art results.
Not currently ranked — collecting fresh signals.
star history
UniMatch proposes a unified architecture that jointly handles optical flow estimation, stereo matching, and depth prediction using cross-attention and correlation mechanisms. The model leverages a transformer backbone with global matching to learn dense correspondences across image pairs. It achieves 1st place on Sintel, Middlebury, and Argoverse benchmarks, outperforming task-specific approaches while using a single model architecture.
Frequently asked
- What is autonomousvision/unimatch?
- A unified transformer-based model for optical flow, stereo matching, and monocular depth estimation achieving state-of-the-art results.
- Is unimatch open source?
- Yes — autonomousvision/unimatch is open source, released under the MIT license.
- What language is unimatch written in?
- autonomousvision/unimatch is primarily written in Python.
- How popular is unimatch?
- autonomousvision/unimatch has 1.4k stars on GitHub.
- Where can I find unimatch?
- autonomousvision/unimatch is on GitHub at https://github.com/autonomousvision/unimatch.