About
We introduce SemTrack, a large and comprehensive dataset containing annotations of the target's semantic trajectory. The dataset contains 6.7 million frames from 6961 videos, covering a wide range of 52 different interaction classes with 115 different object classes spanning 10 different supercategories in 12 types of different scenes, including both indoor and outdoor environments. Our data sources come from YFCC100M, TAO, VIDOR, VIDVRD, HACS, AVA, GOT-10K, ILSVRC2016.
High quality annotated videos
Object Categories
Interaction Categories
Frames in total
Download
Dataset & Toolkit
Download our dataset and toolkit.
Evaluation Code
Start exploring SemTrack.
Publication
If you find our paper helpful, please cite our paper.
Pengfei Wang*, Xiaofei Hui*, Jing Wu*, Zile Yang*, Kian Eng Ong*, Xinge Zhao, Beijia Lu, Dezhao Huang, Evan Ling, Weiling Chen, Keng Teck Ma, Minhoe Hur, Jun Liu
ECCV, 2024
[ArXiv] [Evaluation Code]
Note: Authors with * all have equal contributions.
title={SemTrack: A Large Scale Dataset for Semantic Tracking in the Wild},
author={Wang, Pengfei and Hui, Xiaofei and Wu, Jing and Yang, Zile and Ong, Kian Eng and Zhao, Xinge and Lu, Beijia and Huang, Dezhao and Ling, Evan and Chen, Weiling and Ma, Keng Teck and Hur, Minhoe and Liu, Jun},
booktitle={ECCV},
year={2024}
}