Skip to content

Adaptive Target Tracking Algorithm with Dynamic Weighted Hierarchical Convolutional Neural Network Feature (http://xwxt.sict.ac.cn/CN/abstract/abstract5795.shtml)

Notifications You must be signed in to change notification settings

Lupin1998/ADCF_2019

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ADCF_2019

code for Adaptive Target Tracking Algorithm with Dynamic Weighted Hierarchical Convolutional Neural Network Feature.

1. Intro

  • This project includes source code for first experience in our paper based on PyTorch and usage of CNN features.
  • For models, inlcuding mainstream models from torchvision (PyTorch) in CNNFeat. Models from tensorflow or caffe should perform better on visual tracking, such as (pretrained-models) or (matconvnet).

2. Install

numpy==1.14.5 torch==0.4.0 torchvision==0.4.0 scipy==1.0.0 opencv_python==3.4.2 matplotlib==2.1.2 numba==0.43.1 pillow==6.1.0

3. Usage

  • Download seqs (Visual BenchMark) and save to Sequence. Change file path before running KCF with CNN features. Choose to run different CNN of single conv-layer or coarse-to-fine features in run.py.

    python run.py
    
  • Visualize different CNN feature maps in CNNfeat. Downloads MobileNet_v3 models in (mobilenetv3).

    from PIL import Image
    from CNN_feat.LightWeight_cnn_feat import LightWeight_feat,LightCNN
    net = LightWeight_feat(LightCNN.SqueezeNet.value,'1_1', 1)
    # visualize SqueezeNet-1_1 conv1
    img = Image.open('Seq/boy/img/0001.jpg')
    feat = net.get_cnn_feat(img,True,False,False)
    
  • Test old vision of KCF with hog features in 2015-KCF-DSST. Install 'numba' before if using hog features.

    python run.py
    

About

Adaptive Target Tracking Algorithm with Dynamic Weighted Hierarchical Convolutional Neural Network Feature (http://xwxt.sict.ac.cn/CN/abstract/abstract5795.shtml)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published