…Sequences are annotated with more than 100K coarse and 1M fine-grained action segments, and 18M 3D hand poses. We benchmark on three action understanding tasks: recognition, anticipation and temporal segmentation. Additionally, we propose a novel task of detecting mistakes.
38 PAPERS • 4 BENCHMARKS
…More specifically, the dataset contains 50 hours of annotated videos to localize relevant animal behavior segments in long videos for the video grounding task, 30K video sequences for the fine-grained
15 PAPERS • 2 BENCHMARKS
…A case was composed of kinematic data, a video, semantic segmentation of each frame, and workflow annotation.
3 PAPERS • 6 BENCHMARKS
…We benchmark four foundational video understanding tasks: action recognition, action segmentation, object detection and multi-object tracking.
1 PAPER • NO BENCHMARKS YET
…EPIC-KITCHENS-55), EPIC-KITCHENS-100 has been annotated using a novel pipeline that allows denser (54% more actions per minute) and more complete annotations of fine-grained actions (+128% more action segments
137 PAPERS • 7 BENCHMARKS