InteractTrack

Official Dataset and Codebase for the CVPR 2026 paper:
Interactive Tracking: A Human-in-the-Loop Paradigm with Memory-Augmented Adaptation

This repository contains the dataset, code, and related annotations for InteractTrack, designed to evaluate interactive visual tracking models with a human-in-the-loop paradigm.

🎯 Motivation

Note:The figure above illustrates the core motivation of our work. Unlike traditional one-pass tracking, InteractTrack introduces a Human-in-the-Loop paradigm where user interactions are incorporated to correct tracking failures in complex scenarios.

📥 Download Links

We provide the dataset in a single compressed file InteractTrack.zip. You can download it from the following cloud storage platforms:

• Baidu Wangpan (百度网盘)
  • Link: https://pan.baidu.com/s/1ykDipt00YdHQt-XCaoL0Gw?pwd=INTK
  • Password: INTK
• Google Drive
  • Link: https://drive.google.com/file/d/1GTvVNpWzcMBEfd3D_tgZjphNjSmgHV8h/view?usp=drive_link

Supplementary Annotation Files

The missing switch.txt files are provided as a separate supplement:

• Download link: https://drive.google.com/file/d/1Ahs8LA5BG7rEaVor857c2PnMwoqkSLYb/view?usp=drive_link
• Please place each switch.txt file into its corresponding sequence directory according to the sequence name.

🗂️ Dataset Structure

After downloading and extracting InteractTrack.zip, the directory structure contains all video sequences directly at the root level. Each sequence folder is structured as follows:

InteractTrack/
├── 0001/
│   ├── imgs/                 # Directory containing video frames
│   ├── description.txt       # Textual description of the sequence
│   ├── groundtruth.txt       # Bounding box annotations
│   ├── meta_info.ini         # Meta information (resolution, frame rate, etc.)
│   ├── scene.txt             # Scene context and attributes
│   └── switch.txt            # Interaction/Switch points for human-in-the-loop
├── 0002/
├── Acrobat-17/
├── IceBall_video_02-Done/
├── TUD-Campus/
├── v_-60s86HzwCs_c001/
└── ... (other sequence folders)

🚀 Getting Started

1. Download the dataset from the links above.
2. Extract the contents to your preferred data directory:

   unzip InteractTrack.zip -d /path/to/your/data_directory/

3. Update the data paths in your configuration files to point to the extracted dataset folder.

📝 Citation

If you find this dataset or our paper useful in your research, please consider citing our work:

@inproceedings{interactTrack,
  title={Interactive Tracking: A Human-in-the-Loop Paradigm with Memory-Augmented Adaptation},
  author={Huang, Yuqing and Zeng, Guotian and Yuan, Zhenqiao and He, Zhenyu and Li, Xin and Wang, Yaowei and Yang, Ming-Hsuan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={35166--35176},
  year={2026}
}