Video Segmentation for AI Computer Vision Models
Video
SegmentationI contributed to high-volume video segmentation projects for AI model training on the Mercor platform. My core tasks involved performing precise semantic and instance segmentation on video footage, labeling objects, boundaries, motion, and scenes at the pixel level according to detailed client annotation guidelines. I worked with large datasets of short-to-medium video clips used to train computer vision and Physical AI models. I consistently maintained high accuracy standards (95%+ inter-annotator agreement) through rigorous self-review, following complex edge-case rules, and incorporating feedback from QA rounds. This work directly supported the creation of clean, reliable training data that improves model performance in real world perception tasks.