nxquang-al / yolov1-keras-voc

Keras implementation of YOLOv1 (Tensorflow backend)

yolov1-keras-voc

Keras implementation of YOLOv1 (Tensorflow backend) inspired by allanzelener/YAD2K and qqwweee/keras-yolo3.

Download Yolo weights

Download tiny-yolov1 weights from here.

wget http://pjreddie.com/media/files/yolov1/tiny-yolov1.weights

Convert the model

Convert the Darknet model to a Keras model.

python convert.py cfg/yolov1-tiny.cfg tiny-yolov1.weights tiny-yolov1.hdf5

Train your datasets

I use pascal-v0c2007 to train, you can download the datasets from here.

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

Extract all of these tars into one directory named VOCdevkit

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

Build datasets.

python voc_annotation.py [Your Datasets Path]

Train by own datasets.

python train.py 100 32 [Your Datasets Path]

Detect image

Imagepath is the path of image you want to detect.

python tiny_yolov1.py tiny-yolov1.hdf5 [Your Image Path]