Our code is based on PySOT repository. You may check the original README.md of PySOT.
This code has been tested on centos 7(Ubuntu is also OK), Python 3.6, Pytorch 1.1.0(Pytorch 1.2,1.3,1.4 and 1.5 are also OK, but for Pytorch 1.7.0 and above versions, the testing results will have slight difference), CUDA 10.0. Please install related libraries in INSTALL.MD:
export PYTHONPATH=/path/to/SiamRPN++-RBO:$PYTHONPATH
Download the pretrained model: Google driver or BaiduYun(code:4oh4) and put them into checkpoints directory.
Download testing datasets and put them into testing_dataset directory. Jsons of commonly used datasets can be downloaded from BaiduYun(code: hkfp) or Google driver. If you want to test the tracker on a new dataset, please refer to pysot-toolkit to set test_dataset.
####### GOT-10K dataset ###########
python tools/test.py --dataset GOT-10k --snapshot checkpoints/SiamRPN++-RBO-got10k.pth --config experiments/test/GOT-10k/config.yaml
####### OTB100 dataset ########### python tools/test.py --dataset OTB100 --snapshot checkpoints/SiamRPN++-RBO-general-OTNV.pth --config experiments/test/OTB100/config.yaml
####### TC128 dataset ########### python tools/test.py --dataset TC128 --snapshot checkpoints/SiamRPN++-RBO-general-OTNV.pth --config experiments/test/TC128/config.yaml
####### NFS30 dataset ########### python tools/test.py --dataset NFS30 --snapshot checkpoints/SiamRPN++-RBO-general-OTNV.pth --config experiments/test/NFS30/config.yaml
####### VOT2016 dataset ########### python tools/test.py --dataset VOT2016 --snapshot checkpoints/SiamRPN++-RBO-general-OTNV.pth --config experiments/test/VOT2016/config.yaml
####### UAV123 dataset ########### python tools/test.py --dataset UAV123 --snapshot checkpoints/SiamRPN++-RBO-general-LU.pth --config experiments/test/UAV123/config.yaml
####### LaSOT dataset ########### python tools/test.py --dataset LaSOT --snapshot checkpoints/SiamRPN++-RBO-general-LU.pth --config experiments/test/LaSOT/config.yaml
The testing result will be saved in the results/dataset_name/tracker_name directory.
Download the datasets：
Scripts to prepare training dataset are listed in training_dataset directory.
If you are confused with preparing training datasets, please refers to SiamBAN[https://github.com/hqucv/siamban] for more details about setting training dataset.
We provide cropped images training datasets at Baidu Drive:
- [YOUTUBEBB] url:https://pan.baidu.com/s/1Rsi9nl9fr1bRSMD9MF-emw code:shcv
- [VID] url：https://pan.baidu.com/s/1LoUCnfXmNSStmcr51agK5w code：bwq5
- [DET] url：https://pan.baidu.com/s/1e1iL9hAsOc80A08xOOpL7A code：w2wf
- [COCO] url：https://pan.baidu.com/s/1_goqFHPJvF6cbdJWdcMRAw code: w9xc
- [GOT-10K] url：https://pan.baidu.com/s/1n5wBE3SFbH9eEPBdrDaX7Q code:kwqd
Download pretrained backbones from google driver or BaiduYun (code: 5o1d) and put them into pretrained_models directory.
To train the SiamRPN++-RBO model, run train.py with the desired configs:
cd experiments/train/got10k
cd experiments/train/fulldata
CUDA_VISIBLE_DEVICES=0,1,2,3
python -m torch.distributed.launch
--nproc_per_node=4
--master_port=2333
../../../tools/train.py --cfg config.yaml
We use four RTX 1080ti for training.
We provide the raw tracking results of OTB100, VOT2016, UAV123, NFS30, GOT-10K, TC128 and LaSOT. If you want to evaluate the tracker, please put those results into results directory.
##for example, evalution on the OTB100 dataset
python tools/eval.py --dataset OTB100
The code is implemented based on pysot. We would like to express our sincere thanks to the contributors.
If you use RBO in your work please cite our paper:
@InProceedings{tang_2022_CVPR,
author = {Feng Tang, Qiang Ling},
title = {Ranking-Based Siamese Visual Tracking},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2022}
}