EPIC-KITCHENS
DatasetEPIC-KITCHENS is a large-scale egocentric video benchmark released by the University of Bristol and University of Catania. It was recorded by 32 participants in their native kitchen environments, capturing non-scripted daily cooking activities totaling 100 hours of video, 90K+ action segments, and 45K+ narrated actions. Each action is annotated with a verb and a noun (e.g., "take knife", "open cupboard", "cut carrot"), providing a naturalistic action taxonomy of 113 verbs and 331 nouns. The dataset includes audio, gaze tracking, and accelerometer data for a subset of recordings. EPIC-KITCHENS has become a standard benchmark for egocentric action recognition, anticipation, and video understanding. It is widely used in robotics research for learning manipulation skills from human demonstrations and in vision-language model evaluation.