A model zoo for MATLAB, with the tradeoff numbers up front
A curated index of pretrained networks for MATLAB that lists size, accuracy, and mAP up front, so picking a model becomes arithmetic instead of guesswork.

What it does This is a catalog, not a codebase: an awesome-list-style index of pretrained deep learning models that work in MATLAB, organized by task. Computer vision dominates — classification, detection, segmentation, pose estimation, 3D reconstruction, video, text recognition — but it also covers transformer NLP models, audio tasks like speech-to-text and pitch estimation, lidar point clouds, and robotics motion planning. Every entry comes with model size, a quality metric, and a pointer to either MathWorks documentation or a GitHub repo holding the weights.
The interesting bit The tables make model selection concrete. SqueezeNet at 5.2 MB and 55% accuracy versus NASNet-Large at 332 MB and 81.8% is a tradeoff you can read off two adjacent rows, and the same treatment extends to YOLO v2 through v9 with COCO mAP per variant. It also catalogs domains most zoos ignore — lidar point-cloud networks and motion-planning networks for robot manipulators.
Key highlights
- Roughly 20 image classifiers tabulated by size and accuracy, from
squeezenet(5.2 MB) up to ViT variants (85.59%) - Object detectors spanning YOLO v2–v9, YOLOX, and EfficientDet-D0, with COCO mAP and file sizes per variant
- Beyond vision: speech-to-text, pitch estimation, lidar segmentation, and manipulator motion planning all get entries
- Since R2024a,
imagePretrainedNetworkis the single accessor for the classification models - Each row links to MathWorks docs or a separate GitHub repo with the weights
Caveats
- The repo itself is links and tables — no code, no weights; it’s an index and doesn’t pretend otherwise
- Many entries route into MathWorks documentation, so the practical prerequisite is a MATLAB installation, not just a browser
- Classification accuracies are listed without naming the benchmark dataset (detection is explicitly COCO)
Verdict If you do deep learning in MATLAB, this is the cheat-sheet you’d otherwise assemble yourself from scattered doc pages. If you live in Python, your zoos are already bigger — nothing to see here.
Frequently asked
- What is matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub?
- A curated index of pretrained networks for MATLAB that lists size, accuracy, and mAP up front, so picking a model becomes arithmetic instead of guesswork.
- Is MATLAB-Deep-Learning-Model-Hub open source?
- Yes — matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub is an open-source project tracked on heatdrop.
- What language is MATLAB-Deep-Learning-Model-Hub written in?
- matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub is primarily written in MATLAB.
- How popular is MATLAB-Deep-Learning-Model-Hub?
- matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub has 563 stars on GitHub.
- Where can I find MATLAB-Deep-Learning-Model-Hub?
- matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub is on GitHub at https://github.com/matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub.