This repo is NOT a fully reinplementation of our submitted result(mAP@27.46).
Instead, two models for appearance modeling are included, together with the open-source BAGS model and the full set of code for inference. With this code, you can achieve around mAP@23 with TAO test set (based on our estimation).
Please find instructions for the detection model in tao_detection_release/README.md and the tracking model in tao_tracking_release/README.md
The result for the TAO validation set can be downloaded in TAO_val.
Apache 2.0
If you find this code useful, please cite our arxiv tech report:
@article{Du_2020_TAO,
author = {Fei Du and Bo Xu and Jiasheng Tang and Yuqi Zhang and Fan Wang and Hao Li},
title = {1st Place Solution to {ECCV-TAO-2020:} Detect and Represent Any Object for Tracking},
journal = {arXiv preprint arXiv: 2101.08040},
year = {2021},
url = {https://arxiv.org/abs/2101.08040}
}