fabiotosi92/Awesome-Deep-Stereo-Matching

A curated list of awesome Deep Stereo Matching resources

This is a curated list — a well-organized collection — of the most important papers and resources on deep stereo matching. That's the field where computers estimate depth (3D distance) by comparing two images taken from slightly different viewpoints (like our left and right eyes), using deep learning techniques.

The project is maintained by three researchers from the University of Bologna: Fabio Tosi, Matteo Poggi, and Luca Bartolomei. It was inspired by a similar popular list called awesome-computer-vision.

What it contains

  • Papers: The latest and most influential studies published in top computer-vision conferences and journals, covering all major aspects of deep stereo matching.
  • Categories: All resources are sorted into clear groups to make them easy to browse — for example, foundational works, efficient architectures, handling specific challenges (bad weather, missing depth data), datasets (both real-world and synthetic), software codebases, applications, and more.
  • BibTeX file: A references.bib file with ready-to-use citation entries for every included work.
  • Highlighted key works: A small flag symbol (🚩) marks the absolute most groundbreaking papers.

Related publications

The repository is closely tied to three academic surveys written by the maintainers (published in IJCV and TPAMI), which provide in-depth overviews of the field. They also presented a tutorial on this topic at the CVPR 2024 conference, with slides and extra materials available on a dedicated webpage.

How to contribute and use it

The README includes instructions for submitting a pull request, so researchers can add their latest work. Users are also encouraged to star the repository and cite it in their own work if they find it valuable.

Related

  • Project
    SubhamTyagi/android-ocr

    A privacy-focused Android application that uses Tesseract 5 to perform offline optical character recognition (OCR) in over 120 languages.

  • Project
    talmolab/sleap

    A deep-learning framework for multi-animal pose tracking that uses a human-in-the-loop GUI to rapidly label and quantify animal behavior.

  • Project
    valkia/aramgg_client

    A Windows desktop assistant for League of Legends ARAM mode that provides real-time champion selection advice and in-game Augment recommendations via screen recognition and LCU data.

  • Project
    yzfzzz/depth-detect

    depth-detect is a C++14 framework that combines YOLO object detection with monocular or stereo depth estimation to identify approaching hazards for a bike‑mounted rear‑view mirror. It supports TensorRT and ONNX inference, runs on x86 and Jetson devices, offers three scheduling modes (sync, overlap, stagger), tracks objects with BYTETracker, and sends JSON alerts over TCP. Benchmarks show ~19× speed‑up on Jetson TX2 with TensorRT FP16 and ~13.5 fps in a staggered configuration.