video and text labeling
Video
SegmentationThe Atlas Capture Project is a large-scale global crowdsourcing platform dedicated to collecting and annotating high-quality egocentric (first-person) video data of everyday human activities in natural home and outdoor settings, with the primary goal of training advanced AI and robotics models for physical AI and embodied intelligence. Contributors mainly perform data labeling tasks, which include segmenting videos into meaningful action episodes and applying either dense atomic descriptions (detailed step-by-step hand-object interactions using precise physical verbs) or coarse high-level summaries, while strictly adhering to guidelines on accurate segmentation, object identification, timestamp precision, factual observations, and minimizing hallucinations or intent speculation. Operating with thousands of active remote contributors worldwide and millions of tasks completed, the project maintains rigorous quality standards through detailed playbooks, machine-assisted pre-labeling, regular audits, and consistency checks to produce reliable, production-grade training data.