Skip to content

BirdeeChen/caffe_yolo_v2

 
 

Repository files navigation

YOLO: Real-Time Object Detection (Caffe)

This is the caffe version YOLO V2 ported directly from darknet YOLO

Detection Using A Pre-Trained VOC Model

  • Step 0 BUILD this reporsitory caffe

    git clone https://github.com/quhezheng/caffe_yolo_v2
    cd caffe_yolo_v2
    mkdir build
    cd build
    cmake ..
    make  
    
  • Step 1 Download the trained model from Baidu disk https://pan.baidu.com/s/1jJ9emNW

    cd ..  #back to project root folder
    cd examples/yolo
    mkdir model_voc
    cp DOWNLOAD/PATH/OF/yolo_voc_iter_120000.caffemodel model_voc
    python detect.py
    

    image

    detect.py use CPU do do predition by default, please change the script if want GPU

Train the VOC data

  • Step 0 Build caffe in [YOUR_LOCAL_REPOSITORY_PATH]/build

  • Step 1 Download VOC2007 & 2012 dataset(here)

    cd [YOUR_LOCAL_REPOSITORY_PATH]/data/yolo
    wget http://pjreddie.com/media/files/VOCtrainval_06-Nov-2007.tar
    wget http://pjreddie.com/media/files/VOCtest_06-Nov-2007.tar
    wget http://pjreddie.com/media/files/VOCtrainval_11-May-2012.tar
    wget http://pjreddie.com/media/files/VOC2012test.tar
    tar -xvf VOCtest_06-Nov-2007.tar
    tar -xvf VOCtrainval_06-Nov-2007.tar
    tar -xvf VOCtrainval_11-May-2012.tar
    tar -xvf VOC2012test.tar

    There it will be 'VOCdevkit' folder

  • Step 2 Create lmdb index *.txt files

    python get_list.py
    

    There it will be 'trainval.txt test_2007.txt test_2012.txt' files

  • Step 3 Create lmdb

    ./convert.sh
    

    There it will be lmdb folder has all lmdb

  • Step 4 Download the pre-trained model darknet19_448.conv.23.caffemodel from Baidu disk

    This model is converted directly from darknet darknet19_448.conv.23. It contain trained TOP 23 layers' weight, other layers' weight are initilized by 'xavier'

    cd ../../examples/yolo
    cp DOWNLOAD/PATH/OF/darknet19_448.conv.23.caffemodel ./
    mkdir model_voc
    ./train_voc.sh
    

Packages

No packages published

Languages

  • C++ 79.4%
  • Python 10.2%
  • Cuda 5.7%
  • CMake 2.7%
  • MATLAB 0.9%
  • Makefile 0.6%
  • Shell 0.5%