# Improving Deep Stereo Network Generalization with Geometric Priors

arXiv: https://arxiv.org/abs/2008.11098 · HTML: https://mlsanity.com/paper/2008.11098/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 25 Aug 2020.

## Built on by (most cited first, 8 of 8)

- [Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation](https://mlsanity.com/paper/2203.11483/) (2022)
- [CFNet: Cascade and Fused Cost Volume for Robust Stereo Matching](https://mlsanity.com/paper/2104.04314/) (2021)
- [SLIDE: Single Image 3D Photography with Soft Layering and Depth-aware Inpainting](https://mlsanity.com/paper/2109.01068/) (2021)
- [Uncertainty Guided Adaptive Warping for Robust and Efficient Stereo Matching](https://mlsanity.com/paper/2307.14071/) (2023)
- [Digging Into Uncertainty-based Pseudo-label for Robust Stereo Matching](https://mlsanity.com/paper/2307.16509/) (2023)
- [A Practical Stereo Depth System for Smart Glasses](https://mlsanity.com/paper/2211.10551/) (2022)
- [Rethinking the Key Factors for the Generalization of Remote Sensing Stereo Matching Networks](https://mlsanity.com/paper/2408.07613/) (2024)
- [RGB-Phase Speckle: Cross-Scene Stereo 3D Reconstruction via Wrapped Pre-Normalization](https://mlsanity.com/paper/2503.06125/) (2025)

## Abstract (arXiv)

End-to-end deep learning methods have advanced stereo vision in recent years and obtained excellent results when the training and test data are similar. However, large datasets of diverse real-world scenes with dense ground truth are difficult to obtain and currently not publicly available to the research community. As a result, many algorithms rely on small real-world datasets of similar scenes or synthetic datasets, but end-to-end algorithms trained on such datasets often generalize poorly to different images that arise in real-world applications. As a step towards addressing this problem, we propose to incorporate prior knowledge of scene geometry into an end-to-end stereo network to help networks generalize better. For a given network, we explicitly add a gradient-domain smoothness prior and occlusion reasoning into the network training, while the architecture remains unchanged during inference. Experimentally, we show consistent improvements if we train on synthetic datasets and test on the Middlebury (real images) dataset. Noticeably, we improve PSM-Net accuracy on Middlebury from 5.37 MAE to 3.21 MAE without sacrificing speed.