chasonlee / tensorflow-yolo

tensorflow implementation of 'YOLO : Real-Time Object Detection'(train and test)

tensorflow-yolo

Require

tensorflow-1.0

See https://www.tensorflow.org/install/install_linux#installing_with_virtualenv

opencv

Try pip install opencv-python

If pip install failed, then install it from the source: http://opencv.org/releases.html

numpy

Try pip install numpy

download pretrained model

yolo_tiny: https://drive.google.com/file/d/0B-yiAeTLLamRekxqVE01Yi1RRlk/view?usp=sharing

	mv yolo_tiny.ckpt models/pretrain/ 

Train

Train on pascal-voc2007 data

Download pascal-Voc2007 data

1. Download the training, validation and test data

   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
   

2. Extract all of these tars into one directory named VOCdevkit

   tar xvf VOCtrainval_06-Nov-2007.tar
   tar xvf VOCtest_06-Nov-2007.tar
   

3. It should have this basic structure

   $VOCdevkit/                           # development kit
   $VOCdevkit/VOCcode/                   # VOC utility code
   $VOCdevkit/VOC2007                    # image sets, annotations, etc.
   # ... and several other directories ...
   

4. Create symlinks for the PASCAL VOC dataset

   cd $YOLO_ROOT/data
   ln -s VOCdevkit VOCdevkit2007
   

   Using symlinks is a good idea because you will likely want to share the same PASCAL dataset installation between multiple projects.

convert the Pascal-voc data to text_record file

python tools/preprocess_pascal_voc.py

train

python tools/train.py -c conf/train.cfg

Train your customer data

1. transform your training data to text_record file(the format reference to pascal_voc)

2. write your own train-configure file

3. train (python tools/train.py -c $your_configure_file)

test demo

python demo.py