# NVlabs/FoundationStereo

[CVPR 2025 Best Paper Nomination] FoundationStereo: Zero-Shot Stereo Matching

Repository: https://github.com/NVlabs/FoundationStereo
Canonical: https://ross.abutalabs.com/products/foundationstereo
Homepage: https://nvlabs.github.io/FoundationStereo/
Language: Python
License: NOASSERTION
License Family: other
Last push: 2025-12-19T00:46:24+00:00

## Health v2 (maintenance only)
Score: 47/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 57, release rhythm 35, longevity 44
- inputs: {"age_days": 629, "days_push": 258, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2874, forks 284 (observed 2026-08-28T04:07:27.275761+00:00)

## What it is
FoundationStereo is NVIDIA's official PyTorch implementation of a foundation model for zero-shot stereo depth estimation, published as a CVPR 2025 Oral (Best Paper Nomination). It takes a pair of stereo images and outputs a dense disparity map convertible to metric-scale depth maps or 3D point clouds, with strong cross-domain generalization.

## Use cases
- estimate depth from stereo image pairs
- generate 3D point clouds from stereo cameras
- run zero-shot stereo matching on new domains without fine-tuning
- deploy depth estimation on Jetson with TensorRT
- benchmark stereo depth models on Middlebury and ETH3D
- extract dense disparity maps for robotics perception

## When to choose
- you need state-of-the-art stereo depth estimation that generalizes zero-shot across domains
- you want a pretrained model that works without per-domain fine-tuning
- you need metric-scale depth or point clouds from stereo rigs
- you target embedded deployment via ONNX/TensorRT on Jetson

## When to avoid
- you only have a single monocular camera instead of a stereo pair
- you need real-time inference on limited hardware without the Fast variant
- you require a permissive license for commercial use without checking terms
- you work outside GPU-capable environments

## Facets
- artifact type: library
- maturity: active
- function: computer-vision, image-processing, machine-learning, deep-learning
- domain: computer-vision, deep-learning, robotics, artificial-intelligence
- platform: python, cpp
- tags: stereo-matching, depth-estimation, disparity-map, zero-shot-generalization, foundation-model, 3d-reconstruction, point-cloud, onnx, tensorrt, cvpr-2025, gpu, linux

## Member repositories
- NVlabs/FoundationStereo (main) score 47

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:27.275761+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:35:48.347599+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/FoundationStereo (fetched 2026-08-28T04:07:27.275761+00:00, sha 08a870ed8dab)
  - homepage: https://nvlabs.github.io/FoundationStereo/ (fetched 2026-08-29T09:51:18.566523+00:00, sha d828e35c12c7)
- Data as of 2026-08-30T08:39:29.467469+00:00.