CoolFace
Apppublic

hugging-apps/fast-foundationstereo

sourceHugging Faceupdated 3mo agoView on Hugging Face
0likes
App README

Fast-FoundationStereo

Real-time zero-shot stereo disparity estimation using nvidia/c-fast-foundationstereo.

Upload a rectified binocular stereo image pair (left + right) to estimate a dense disparity map. The model is a 14.6M-parameter transformer-based stereo foundation model from NVIDIA (CVPR 2026) that achieves real-time frame rates while maintaining strong zero-shot generalization.

Usage

  1. 1.Upload or select a rectified left stereo image.
  2. 2.Upload or select the corresponding rectified right stereo image.
  3. 3.Click "Estimate Disparity" to get a color-mapped disparity visualization and raw 16-bit disparity map.

Notes

  • —Input images must be rectified (epipolar lines are horizontal, no lens distortion).
  • —Do not swap left and right images — the left image should be from the left-side camera.
  • —For best results use PNG images with no lossy compression.
  • —The model performs best with image width < 1000 pixels.

References