Atlas Capture
Atlas Capture is a video action segmentation and annotation project focused on labeling observable human-object interactions to train AI and computer vision systems. The work involves reviewing short video clips, identifying clear action boundaries, and applying standardized verb-object labels such as pick, place, move, adjust, rotate, and wipe while maintaining consistent segmentation granularity throughout each task. The project requires careful handling of stacked objects, continuous manipulations, and simultaneous actions, with strict adherence to labeling guidelines, precise boundary detection, and quality control standards to ensure accurate and reliable structured training data.