Evaluation Data (Video Recognition)
Kinetics-400
#Evaluation Metrics-Acc@1, Acc@5
Data description:
Kinetics-400 is a large-scale video action recognition dataset released by the DeepMind team in 2017. It is designed for training and evaluating video action recognition models and is currently widely used in research in this field. Kinetics-400 significantly expands the data scale and diversity of video action recognition datasets. Its video data is sourced from YouTube and includes 400 different action categories, covering most of the daily life scenarios of humans. Each video clip is approximately 10 seconds long to ensure that each video mainly showcases a specific action.
The dataset includes a total of over 300,000 video samples, with nearly 240,000 videos in the training set and 20,000 videos each in the validation set and the test set. Since the annotations of the original test set are not publicly available, the original validation set is used as the test set.
Dataset structure:
Amount of source data:
The dataset is divided into auxiliary training set (240,435), validation set (19,795), and test set (19,795).
Amount of test data:
All 19,795 test examples from the source dataset test set.
Data detail:
| KEYS | EXPLAIN |
|---|---|
| videos | videos used as input |
| labels | Category label of the video |
Sample of source dataset:
video：
label：
brushing hair
Paper citation:
@article{kay2017kinetics,
title={The kinetics human action video dataset},
author={Kay, Will and Carreira, Joao and Simonyan, Karen and Zhang, Brian and Hillier, Chloe and Vijayanarasimhan, Sudheendra and Viola, Fabio and Green, Tim and Back, Trevor and Natsev, Paul and others},
journal={arXiv preprint arXiv:1705.06950},
year={2017}
}