NVlabs/Fast-FoundationStereo

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1432 stars · 143 forks observed · 2026-08-28

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

Fast-FoundationStereo is NVIDIA's official PyTorch implementation of a real-time zero-shot stereo matching model family, accepted to CVPR 2026. It distills and prunes the FoundationStereo model to run over 10x faster while closely matching its zero-shot accuracy for disparity/depth estimation from stereo image pairs.

Use cases

• estimate depth from stereo image pairs in real time
• run zero-shot stereo matching without per-domain fine-tuning
• compute disparity maps for robotics or AR applications
• trade off speed versus accuracy with model variants
• benchmark real-time stereo matching methods

When to choose

• you need real-time stereo depth estimation with strong generalization to unseen domains
• you want a drop-in faster alternative to FoundationStereo for online inference
• you have a CUDA GPU and want state-of-the-art zero-shot disparity results

When to avoid

• you need the absolute highest offline accuracy and latency does not matter (use FoundationStereo instead)
• you need CPU-only inference
• you need a production-supported library with a stable API and permissive license

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing computer-vision machine-learning deep-learning robotics python stereo-matching depth-estimation disparity zero-shot real-time knowledge-distillation neural-architecture-search research-code cvpr gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVlabs/Fast-FoundationStereo")