Implementation code for eccv 2018:
[1] X. Dong and J. Shen
Triplet loss in siamese network for object tracking,
ECCV, pp. 459-474, 2018
========================================================================
Any comments, please email: xingping.dong@gmail.com
shenjianbingcg@gmail.com
This software was developed under Ubuntu 14.04 with matlab 2017a.
If you use this software for academic research, please cite the following paper:
@inproceedings{dong2018triplet,
title={Triplet loss in siamese network for object tracking},
author={Dong, Xingping and Shen, Jianbing},
booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
pages={459--474},
year={2018}
}
[**Prerequisites**]
Cuda 7.5
cuDNN v5.1
Matconvnet v1.0-beta20
Matlab 2017a
[**Training**] After you finish all Prerequisites, you can train our network step by step.
1. Perpare dataset: (same instructions in SiamFC:'Fully-Convolutional Siamese Networks for Object Tracking' ) Follow these [step-by-step instructions] in ./ILSVRC15-curation/README.md
2. Setup the paths in ./training/env_paths_training.m and startup.m
3. Run the script for training:
./training/run_experiment_tri.m
4. After training, you can select the epoch with the lowest val errdisp according ./training/triplet_gray/net-train.pdf (net-epoch-1.mat in this test code). Then this model can be used for tracking in next step.
[**Tracking**]
1. You can directly run './tracking/demo.m' as a demo to test our algrithom with the pre-trained model ('./models/triplet.net.mat'). You can also try your trained model by modifying ./tracking/run_tracker.m and env_paths_tracking.m.
Note: some codes are borrow from "SiamFC". Thanks for all authors of SiamFC.