Latent Assisted Label¶
Latent Assisted Label is an AI-powered assisted labeling tool that accelerates object detection and dataset annotation. It is available as a FiftyOne plugin with a standalone CLI launcher.
It combines Vision-Language Models (VLMs) with the Segment Anything Model (SAM) for comprehensive object discovery, uses embedding-based clustering with majority-vote labeling and LDA classification to propagate labels across similar objects, tracks and labels identical objects across video sequences, and allows quick editing and pruning of detections with safety toggles.
Latent Assisted Label reduces the manual effort needed to annotate large collections, addresses class imbalance with clustering, speeds up repetitive labeling in video, and gives annotators faster ways to refine and correct outputs.
If you're an ML engineer building or refining object detection datasets, part of an annotation team that needs faster workflows without sacrificing accuracy, or a researcher working with vision-language models, SAM, or semi-supervised labeling techniques, you should use Latent Assisted Label.
In this documentation¶
-
Start here for step-by-step instructions for installing and getting started with Latent Assisted Label
-
Guides that cover key operations and common tasks
-
Reference documentation for Latent Assisted Label
-
Learn more about Latent Assisted Label