donghaiwang / VisualTracking_DRL

使用深度强化学习进行多目标跟踪

Multi-Object Tracking Use Deep Reinforcement Learning

• ADNet: Action-Decision Networks for Visual Tracking with Deep Reinforcement Learning [Paper]
• [Project Page]
• MDP_Tracking: Learning to Track: Online Multi-Object Tracking by Decision Making
• [Project Page]
• Email: [e-mail] [main page]

Install

• Test on (ubuntu 14.04 and ubuntu 16.04) 64bit, MATLAB 2017a and MATLAB 2018a, use Cuda-8.0 and NVIDIA GTX 1080 TI.
• Run adnet_compile.m to compile MatConvNet.
• modify train data path: utils/get_benchmark_path.m
• 'models/imagenet-vgg-m-conv1-3.mat' copy from [MDNet]

Run tracking

• run adnet_demo.m.

Other

• [[ADNet]] 0.646 0.880
• [DRLT] 0.635
• [DRL-IS] 0.671 0.909
• [MDNet] 0.678 0.909
• [SANet] 0.661 0.913
• [ECO] 0.691 0.910
• [CFNet]

Prospect

• [Biology of Vision]
• [Active Object Tracking]
• [Visual Navigation]