OVT-B-Dataset

This repository contains the download link and usage instructions for the new dataset.

News

🔥 [26/09/24] We are pleased to announce that our paper has been accepted to NeurIPS 2024 Datasets and Benchmarks Track.

🔥 [02/09/24] We have uploaded the source code of several trackers implemented on our benchmark.

Download

You can download the dataset from the following link: BaiDuYun (8yy3) and GoogleDrive.

Description

This dataset OVT-B can be used as a new benchmark to the research of OVMOT.

Usage

Instructions on how to use the dataset:

1. Download the dataset and annotation.
2. Extract the files.
3. Copy the CLASS, base_id, and novel_id from ovtb_classname.py and add them to the classname.py file under the roi_head folder of the ov detector.
4. Modify the data_root in the configs to the path where the OVT-B folder is located. Change ann_file to the path of ovtb_ann.json, img_prefix to data_root+'OVT-B', and prompt_path to the path of ovtb_class.pth.
5. Then test/evaluate by TAO-type/COCO-type dataset eval tools/codes.

Organization

├── OVT-B
│   ├── AnimalTrack
│   │   ├── subdir
│   │   │   ├── img.jpg
│   │   │   └── ...
│   ├── GMOT-40
│   ├── ImageNet-VID
│   ├── LVVIS
│   ├── OVIS
│   ├── UVO
│   ├── YouTube-VIS-2021
├── ovtb_ann.json
├── ovtb_class.pth
├── ovtb_classname.py
├── ovtb_prompt.pth
└── OVTB-format.txt

Data sample illustration

Baseline Method

For detailed information on the baseline method, please refer to the OVTrack.

Citation

If you use this dataset in your research, please cite it as follows:

@article{haiji2024ovtb,
  title={OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking},
  author={Liang, haiji and Han, Ruize},
  journal={arXiv preprint arXiv:2410.17534},
  year={2024}
}

Acknowledgement

• Thanks TETA for providing the evaluation code.
• Thanks DetPro for providing the pytorch reimplementation of VilD.
• Thanks OVTrack for providing the baseline of OVMOT.
• Thanks MMTracking for providing the code of OC-SORT, ByteTrack and StrongSORT.