SeasonStereo — checkpoint

Final SeasonStereo weights: MonSter++ fine-tuned for diachronic stereo matching on multi-date satellite imagery (pseudo-GT disparity + photometric + edge-aware smoothness losses).

• Code
• Data
• Project page
• Paper

Usage

hf download Alvaritox/seasonstereo season-stereo-final.pth --local-dir checkpoints
bash season_stereo/run_evaluation.sh

The checkpoint is loaded by the training/evaluation code in the GitHub repository; see its README.md for environment setup and the expected data/ layout.

Results

Average altitude MAE of 1.05 m on the diachronic test set, matching the LiDAR-supervised variant.

Citation

@inproceedings{diaz-laureano2026seasonstereo,
  title     = {SeasonStereo: Robust Dense Stereo Matching for Multi-Date Satellite Imagery via Generative AI},
  author    = {D{\'\i}az-Laureano, {\'A}lvaro and Mar{\'\i}, Roger and Masquil, El{\'\i}as and Arias, Pablo and Facciolo, Gabriele},
  booktitle = {Proceedings of the TerraBytes II Workshop at the European Conference on Computer Vision (ECCV)},
  year      = {2026}
}