Skip to content

Dataset · lightly-ai

37,455 first-person kitchen action video clips

lightly-ai/epic-kitchens-100-clips

Trimmed clips from EPIC-KITCHENS-100 with text descriptions, verbs, and objects.

What's inside

The dataset collects 37,455 short egocentric video clips from the popular EPIC-KITCHENS-100 set, where each clip corresponds to a single voiced action. Along with the video, tabular metadata with text descriptions, verb/noun breakdowns, and a tool to find discrepancies between clips and captions are available. Prepared by the Lightly team for exploration in LightlyStudio.

What you can do with it

  • Training action recognition models on first-person perspective videos
  • Training multimodal video retrieval systems using text queries
  • Cleaning video datasets and identifying annotation errors using LightlyStudio
Size
10K–100K rows
Downloads this month
28.1K
Likes
4

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("lightly-ai/epic-kitchens-100-clips")