Spaces:

---

Kristen-Z

/

MegaFlow-demo

like 6

Running on Zero

We are excited to share the interactive demo for MegaFlow, a collaborative open-source project from ETH Zürich, Microsoft, and the University of Tübingen. MegaFlow introduces a unified framework that achieves state-of-the-art, zero-shot performance for large displacement optical flow and point tracking.

We have fully open-sourced our code, weights, and evaluation scripts, and we built this Space so the computer vision community can easily test our model on their own images and videos.

Currently, the dense matching computations and feature extractions required for high-resolution optical flow are too heavy for the standard CPU tier, leading to timeouts and a poor interactive experience. A GPU grant would allow anyone to test our model in real-time directly in their browser, removing the barrier of needing local hardware setups.

We would be incredibly grateful for the Hugging Face team's support to make this open-science demo fast and accessible to everyone!

Resources:

• Demo Space: https://huggingface.co/spaces/Kristen-Z/MegaFlow-demo
• Project Page: https://kristen-z.github.io/projects/megaflow/
• Paper: https://arxiv.org/abs/2603.25739
• GitHub: https://github.com/cvg/megaflow