lightly-studio

Why we included this project

The slowest part of many computer vision projects is assembling a clean, well-labeled dataset, and LightlyStudio focuses on that exact bottleneck. It installs as a local app that opens in your browser. You load your images or videos, and an embedding plot shows how they group into clusters, so you can keep only the examples your model actually needs to train on. For labeling, there is a built-in annotation editor with optional plugins such as SAM that automate part of the work. Everything runs on your own hardware, which makes the tool a good match for datasets you cannot send to a cloud labeling service. Teams building YOLO or COCO-based models can handle the whole data-prep workflow in one place, then export the result in the format their pipeline expects.