tensorflow YOLO-v1

• this is a study project, which is not exactly the same as the original paper in the Network structure,we use vgg16 achieve 65.34% map in the voc2007test.

• The purpose that I write the yolov1 to study,and it achieve a good performance.

Train on voc2007+2012

model	map@voc2007test
VGG16	65.34%
VGG19	66.12%
Resnet50	65.23%

1.requirement

• opencv
• tensorflow 1.8
• numpy

2.prepare dataset

1. download voc2007 and voc2012 dataset

2. unzip dataset as following

   • VOCdevkit
     • VOC2007
       • Annotations
       • ImageSets
       • JPEGImages
       • SegmentationClass
       • SegmentationObject
     • VOC2012

3. python proprecess_pasval_voc.py

3.training

python train.py \\  
	--net {Vgg16||Vgg19||resnet50}
	--gpu 0

4.predict and eval

python predict --net {Vgg16||Vgg19||resnet50}
python eval.py

result