Fast-FoundationStereo brings zero-shot stereo matching to real-time speeds
rsasaki0109 · x · 2026-07-21
Fast-FoundationStereo: real-time zero-shot stereo matching
The paper introduces Fast-FoundationStereo, a family of architectures that aims to keep the strong zero-shot generalization of stereo foundation models while making them fast enough for real-time use.
What it does
- Combines knowledge distillation to compress a hybrid backbone into an efficient student model.
- Uses blockwise neural architecture search to discover cost-filtering designs under latency constraints.
- Applies structured pruning to remove redundancy in the iterative refinement module.
- Adds an automatic pseudo-labeling pipeline that curates 1.4M in-the-wild stereo pairs to supplement synthetic data.
Reported result
- The model runs over 10× faster than FoundationStereo while staying close to its zero-shot accuracy.
- The authors claim this sets a new state of the art among real-time stereo methods.
More from Research
- Research finds memory compression makes AI agents drop safety rules and hit 59% violations — gerardsans · 2026-07-22
- DriftWorld claims a world model that runs at 30+ FPS and trains on 1–2 GPUs — du_yilun · 2026-07-22
- Why a 1GW Chinese AI data center may be plausible after all — teortaxesTex · 2026-07-22
- Chinese AI labs are now treating distillation obfuscation as the top research topic — pmddomingos · 2026-07-22
- RSS launches under OMSF to push structural biology data modeling at scale — MoAlQuraishi · 2026-07-22
- enFoldX turns AlphaFold3 ensemble noise into a TCR–peptide–MHC predictor — quaidmorris · 2026-07-22