MRSC-V: Moroccan Road-Sign Crops from driving Videos (images, restricted)

Crop images of 2,081 traffic signs cut from public YouTube driving videos filmed on 9 Moroccan drives (Casablanca, Rabat, Fes, Marrakech, Midelt–Fes, Fes–Ifrane and motorways). The images are PNG files at their original size. The crops are labelled with 70 Morocco-relevant TS-1M classes, plus “other sign” for signs outside these classes. There are three splits, each from different drives: training: 868 crops, 3 drives development: 696 known + 70 unknown, 4 drives (used for model selection) final test: 423 known + 24 unknown, 2 drives (labelled blind and evaluated once) Access: the images are restricted to non-commercial research and teaching use, because the video frames belong to their authors (listed in videos.csv). Access is granted on request. Reviewers of the associated manuscript will be granted access. Contents: images/<split>/<class>/*.png, annotation files with labels and box coordinates, the video list, per-class statistics, SHA-256 checksums of every image and the terms of use (TERMS_OF_USE.txt). The labels, boxes and statistics alone are openly available (CC BY 4.0) at https://doi.org/10.5281/zenodo.23922220. Code and models: https://github.com/badraitmessaad-ing/tinyml-moroccan-signs. Made for: B. Ait Messaad, S. Motahhir, “TinyML traffic-sign classification for Moroccan roads under cross-country domain shift: local adaptation and microcontroller deployment”, Digital Engineering (submitted), 2026.