matteo-dunnhofer / vot-kd-rl

Official code implementation of the papers "Tracking-by-Trackers with a Distilled and Reinforced Model" (ACCV 2020) and "Visual Tracking by means of Deep Reinforcement Learning and an Expert Demonstrator" (ICCVW 2019).

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Tracking-by-Trackers

PWC

PWC

PWC

Official implementation of the tracking-by-trackers framework proposed in the paper "Tracking-by-Trackers with a Distilled and Reinforced Model" (presented at the Asian Conference on Computer Vision 2020).

In particular, the repository contains the official implementation of the TRAS, TRAST, TRASFUST (ACCV 2020), and A3CT, A3CTD (ICCVW 2019) trackers, including trained models, and raw results.

TRAS, TRAST, TRASFUST

[Paper] [Qualitative results] [Raw results] [Pretrained Model]

drawing

A3CT, A3CTD

[Paper] [Qualitative results] [Pretrained Model]

drawing

Installation

Code has been developed and tested on Ubuntu 18.04 with Python 3.6, PyTorch 1.4.0, and CUDA 10.

Clone the GIT repository.

git clone https://github.com/dontfollowmeimcrazy/vot-kd-rl.git

Set paths to checkpoint.

Download the pretrained weights file from here, put wherever you want in your file system, and set the variable CKPT_PATH variable (contained in file track/config_track_accv.py) to point to it.

Test

Set path to benchmark datasets.

In the file track/config_track_accv.py set the variable DATA_PATH to the location of the benchmark datasets (if you do not have them on your machine they will be automatically downloaded by the toolkit at the specified DATA_PATH).

Run the script track/run_test.py by specifing the tracker with the --tracker option and the dataset with the --dataset.

cd track
python run_test.py --tracker TRAS --dataset OTB2015  

For the TRAST, TRASFUST, and A3CTD trackers you either need to: + provide the implementation of "teacher" trackers according to the GOT-10k toolkit tracker class definition, and initialize them in lines 103 and 232 of the track/Trackers.py file. + use the precomputed results of the "teacher" trackers. In this case, you have to specify the variable RESULTS_PATH which should point to the results folder used by the GOT-10k toolkit. Results should be organised as given by the toolkit (e.g. benchmark-name/tracker-name/*.txt). Some examples are given in the folder trackers/results/* for the OTB2015 results of ECO, MDNet, and SiamFC trackers.

Training

Training code will be released soon!

References

If you find this work useful please cite

@InProceedings{Dunnhofer_2020_ACCV,
	author    = {Dunnhofer, Matteo and Martinel, Niki and Micheloni, Christian},
	title     = {Tracking-by-Trackers with a Distilled and Reinforced Model},
	booktitle = {Proceedings of the Asian Conference on Computer Vision (ACCV)},
	month     = {November},
	year      = {2020}
}

@InProceedings{Dunnhofer_2019_ICCVW,
	author 	  = {Dunnhofer, Matteo and Martinel, Niki and Luca Foresti, Gian and Micheloni, Christian},
	title 	  = {Visual Tracking by Means of Deep Reinforcement Learning and an Expert Demonstrator},
	booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
	month 	  = {Oct},
	year 	  = {2019}
}   

Acknowledgements

Code for evaluation was borrowed from the GOT-10k toolkit.

About

Official code implementation of the papers "Tracking-by-Trackers with a Distilled and Reinforced Model" (ACCV 2020) and "Visual Tracking by means of Deep Reinforcement Learning and an Expert Demonstrator" (ICCVW 2019).

License:GNU General Public License v3.0


Languages

Language:Python 100.0%