qianqianwang68/omnimotion

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2268 stars · 132 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

OmniMotion is a PyTorch implementation of the ICCV 2023 paper 'Tracking Everything Everywhere All at Once', which tracks every point in a video using a quasi-3D neural representation. It fits a per-video model that yields dense, global, long-range motion correspondences, depth, and occlusion information.

Use cases

• track every point in a video across all frames
• compute dense long-range motion correspondences in video
• estimate pseudo-depth and occlusion for video sequences
• research video motion tracking with neural representations
• visualize point trails and correspondences in video
• reproduce ICCV 2023 paper results on video tracking

When to choose

• you need dense, globally consistent point tracking across an entire video
• you want depth and occlusion estimates alongside motion tracking
• you are doing computer vision research on video motion and can afford GPU training time

When to avoid

• you need real-time tracking - training takes 8-13 hours on a high-end GPU per sequence
• you want a plug-and-play pretrained tracker without per-video optimization
• you lack a CUDA-capable GPU or are on Windows/macOS without setup effort

Facets

library · maturity maintenance

computer-vision machine-learning deep-learning image-processing video-processing computer-vision machine-learning deep-learning python optical-flow point-tracking neural-representation quasi-3d research-code pytorch iccv-2023 video linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="qianqianwang68/omnimotion")