// the find
SkalskiP/make-sense
Free to use online tool for labelling photos. https://makesense.ai
A browser-only tool for drawing bounding boxes, polygons, lines, and points on images, aimed at people assembling small object-detection or pose-estimation datasets. It runs on React/Redux with TensorFlow.js doing the AI-assisted labeling (YOLOv5, SSD, PoseNet) entirely client-side, so there's no backend and no image upload.
Images never leave the browser — the AI suggestions run on TF.js locally, which is a real privacy property, not a marketing line, and it also means zero server cost to run. It covers the actual annoying part of dataset prep: YOLO, VOC XML, COCO JSON, and VGG JSON import/export are all wired up for the common label types. The YOLOv5 integration via a custom yolov5js port lets you load your own trained models, not just a fixed pretrained set.
No commits or releases since August 2024 — over two years stale as of now, which matters a lot for a project glued to TF.js and a hand-rolled YOLOv5.js port, since both move fast and this isn't tracking them. Their own format matrix admits several combinations are unimplemented (e.g. polygon import from CSV, point export to YOLO), so 'supports YOLO/COCO/VOC' overstates what actually works per label type. It's GPL-3.0, which rules it out for anyone wanting to fork it into a closed-source internal tool. It's single-user with no review or versioning workflow, so it stops being useful the moment more than one person needs to label the same dataset.