Sensor Fusion Labeling

Self-driving models learn from hand-labeled data in which every frame pairs a lidar point cloud with camera images. This web app lets annotators box objects in 3D, check each box from top, side and front views, and label the matching camera images alongside. Throughput was the hard part: frames are prefetched and parsed in Web Workers and cached in IndexedDB, so long sequences play through without load stalls.