This is the repo for paper "Discriminative and Robust Online Learning for Siamese Visual Tracking" [paper] [results], presented as poster at AAAI 2020.
The proposed Discriminative and Robust Online Learning (DROL) module is designed to work with a variety of off-the-shelf siamese trackers. Our method is extensively evaluated over serveral mainstream benchmarks and is believed to induce a consistant performance gain over the given baseline. The model includes but not limited to, as paper evaluated:
The corresponding offline-trained models are availabe at PySOT Model Zoo.
- Please find installation instructions for PyTorch and PySOT in
INSTALL.md. - Add DROL to your PYTHONPATH
export PYTHONPATH=/path/to/drol:$PYTHONPATHDownload models in PySOT Model Zoo and put the model.pth to the corresponding directory in experiment.
cd experiments/siamrpn_r50_l234_dwxcorr
python -u ../../tools/test.py \
--snapshot model.pth \ # model path
--dataset VOT2018 \ # dataset name
--config config.yaml # config fileassume still in experiments/siamrpn_r50_l234_dwxcorr_8gpu
python ../../tools/eval.py \
--tracker_path ./results \ # result path
--dataset VOT2018 \ # dataset name
--num 1 \ # number thread to eval
--tracker_prefix 'model' # tracker_name-
For
DROL-RPN, we have seperate config file thus each own experiment file folder forvot/votlt/otb/others, wherevotis used forVOT-20XX-baselinebenchmark,votltforVOT-20XX-longtermbenchmark,otbforOTB2013/15benchmark, andothersis default setting thus for all the other benchmarks, including but not limited toLaSOT/TrackingNet/UAV123. -
For
DROL-FC/DROL-Mask, only experiments onvot/otbare evaluated as described in the paper. Similar to the repo ofPySOT, we use config file forvotas default setting. -
Since this repo is a grown-up modification of PySOT, we recommend to refer to PySOT for more technical issues.
- Jinghao Zhou, Peng Wang, Haoyang Sun, 'Discriminative and Robust Online Learning For Siamese Visual Tracking', Proc. AAAI Conference on Artificial Intelligence (AAAI), 2020.