Dataset Card for LaSOT

Dataset Summary

Large-scale Single Object Tracking (LaSOT) aims to provide a dedicated platform for training data-hungry deep trackers as well as assessing long-term tracking performance.

This repository contains the conference version of LaSOT, published in CVPR-19 (LaSOT: A High-quality Benchmark for Large-scale Single Object Tracking).

LaSOT is featured in:

Large-scale: 1,400 sequences with more than 3.5 millions frames
High-quality: Manual annotation with careful inspection in each frame
Category balance: 70 categories, each containing 20 sequences
Long-term tracking: An average video length of around 2,500 frames (i.e., 83 seconds)
Comprehensive labeling: Providing both visual and lingual annotation for each sequence

For the new subset (15 categories with 150 videos) in extended journal version (commonly referred to as LaSOT_ext), visit this repo.

Download

You can download the whole dataset via the huggingface_hub library (guide):

from huggingface_hub import snapshot_download
snapshot_download(repo_id='l-lt/LaSOT', repo_type='dataset', local_dir='/path/to/download')

Alternatively, download the videos of a specific category manually from this page.

LaSOT is also distributed through several cloud storage services (currently only OneDrive):

As a single zip file: OneDrive
As one zip file per category: OneDrive or Baidu Pan

Setup

Unzip all zip files and the paths should be organized as following:

├── airplane
│   ├── airplane-1
│   ...
├── basketball
...
├── training_set.txt
└── testing_set.txt

Evaluation Metrics and Toolkit

See the homepage for more information.