Labelme

LabelMe is the rare annotation tool that treats privacy as a feature, not an afterthought. Everything runs offline—your images never leave the machine. For medical imaging, satellite data, or any project under NDA, that's the whole ballgame. The AI assistance is genuinely useful: click to segment with SAM2/SAM3, or type a prompt for YOLO-World to detect objects. In practice, we'd reach for this when annotating a few thousand images for a custom YOLO model—the Toolkit exports directly to YOLO format, and the CLI batch tools handle validation. Where it bites: there's no multi-user real-time collaboration. If your team of five needs to annotate the same dataset simultaneously, you'll pass files around or adopt a cloud tool. The free version requires Python and manual model weight downloads, which isn't for everyone. Compared to Roboflow (which has a free tier but uploads to cloud), LabelMe's one-time pricing wins for long-term ownership—Pro ($79 once) versus Roboflow at $149/year feels like a steal after year two. The recent v6.3.1 crash fix and AI deduplication show active maintenance. One caveat: '1 year of updates' on Starter/Pro means after 12 months you keep the app but stop getting new features (unless you buy Lifetime). That's fair but worth noting. Bottom line: if you value privacy and own your data, LabelMe is a strong pick. It's not for everyone, but for those it fits, it's a gem.