nv-tlabs/vipe

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

2092 stars · 170 forks observed · 2026-08-28

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

ViPE is an open-source video processing engine from NVIDIA that estimates camera intrinsics, camera motion, and dense near-metric depth maps from unconstrained raw videos. It supports pinhole, wide-angle, and 360-degree panorama footage and combines SLAM-style bundle adjustment with monocular depth networks.

Use cases

• annotate camera poses and depth maps from raw videos
• estimate camera intrinsics and motion from selfie or dashcam footage
• generate 3D annotations for spatial AI training data
• process 360-degree panorama videos for 3D reconstruction
• run SLAM-style pose estimation on unconstrained video
• recover metric depth from monocular video

When to choose

• you need automated 3D annotations (poses, intrinsics, depth) from in-the-wild videos
• you work with diverse camera models including wide-angle and panoramas
• you want a GPU-accelerated pipeline with CUDA optimizations
• you are building spatial AI or robotics datasets from video

When to avoid

• you need real-time on-device SLAM on embedded hardware
• your videos are static images rather than video sequences
• you require a permissively licensed codebase without non-commercial components (Unik3D part is BY-NC-SA 4.0)
• you need a lightweight CPU-only tool

Facets

library · maturity active

computer-vision machine-learning video-processing gpu-computing computer-vision robotics artificial-intelligence python cli slam camera-pose-estimation depth-estimation 3d-perception spatial-ai nvidia bundle-adjustment panorama video-annotation video linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="nv-tlabs/vipe")