What it does

Yolo_mark is a keyboard-driven desktop GUI for drawing and editing bounding boxes on images, producing the exact text files YOLO v2/v3 expects for training. It runs on Windows and Linux, speaks OpenCV 2.x or 3.x, and includes a video-to-frames extractor so you can mine stills from footage without leaving the repo.

The interesting bit

The whole thing is essentially a thin OpenCV wrapper—C++ code, MSVS solution, cmake file, done. The value is in the tight coupling to the Darknet ecosystem: it ships with sample .cfg, .data, .names, and even a .cmd file that shows how to kick off training once your thumbs are sore from labeling. The “track objects” shortcut (o) and copy-previous-mark (p) suggest someone actually used this for repetitive frames.

Key highlights

• Native Windows (MSVS 2013/2015) and Linux builds; no Python stack required
• Keyboard-heavy workflow: arrow keys to flip images, r to delete, 0-9 for class IDs, p to clone the last box
• Built-in video frame extractor (cap_video mode) dumps every Nth frame for batch labeling
• Ships with working YOLO v3 config templates and a training command reference
• Outputs YOLO-format .txt files directly—no conversion scripts needed

Caveats

• README still references MSVS 2013/2015 and OpenCV 2.x; the build instructions feel circa-2017
• Windows setup requires hand-editing library paths in Visual Studio—no vcpkg, no Conan
• Only .jpg mentioned for custom images; support for other formats is unclear

Verdict

Grab this if you’re already in the Darknet/YOLO v2-v3 pipeline and want a fast, offline labeler without installing a Python annotation suite. Skip it if you’ve moved on to YOLOv8+, need team collaboration, or want modern formats like COCO JSON export.