SiamPW-RBO

Our code is based on SiamBAN repository. Thansk for their contributions.

1. Environment setup

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 before running this code:

python setup.py build_ext --inplace

Add SiamPW-RBO to your PYTHONPATH

#export PYTHONPATH=/path/to/SiamPW-RBO:$PYTHONPATH

2. Test

Download the pretrained model: Google driver 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/SiamPW-RBO-got10k.pth --config experiments/test/GOT-10k/config.yaml

####### OTB100 dataset ########### python tools/test.py --dataset OTB100 --snapshot checkpoints/SiamPW-RBO-general-OT.pth --config experiments/test/OTB100/config.yaml

####### TC128 dataset ########### python tools/test.py --dataset TC128 --snapshot checkpoints/SiamPW-RBO-general-OT.pth --config experiments/test/TC128/config.yaml

####### NFS30 dataset ########### python tools/test.py --dataset NFS30 --snapshot checkpoints/SiamPW-RBO-general-LUVN.pth --config experiments/test/NFS30/config.yaml

####### VOT2016 dataset ########### python tools/test.py --dataset VOT2016 --snapshot checkpoints/SiamPW-RBO-general-LUVN.pth --config experiments/test/VOT2016/config.yaml

####### UAV123 dataset ########### python tools/test.py --dataset UAV123 --snapshot checkpoints/SiamPW-RBO-general-LUVN.pth --config experiments/test/UAV123/config.yaml

####### LaSOT dataset ########### python tools/test.py --dataset LaSOT --snapshot checkpoints/SiamPW-RBO-general-LUVN.pth --config experiments/test/LaSOT/config.yaml

The testing result will be saved in the results/dataset_name/tracker_name directory.

3. Train

Prepare training datasets

Download the datasets：

• VID
• YOUTUBEBB
• DET
• COCO
• GOT-10K
• LASOT

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.

Download pretrained backbones

Download pretrained backbones from google driver or BaiduYun (code: 5o1d) and put them into pretrained_models directory.

Train a model

To train the SiamPW-RBO model, run train.py with the desired configs:

training got10k model

cd experiments/train/got10k

training general model

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.

4. Evaluation

We provide the raw tracking results of OTB100, VOT2016, UAV123, NFS30, GOT-10K,TC128 and LaSOT at . 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

5. Acknowledgement

The code is implemented based on pysot and siamban. We would like to express our sincere thanks to the contributors.

6. Cite

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}
}