darknet_yolov3

Pytorch darknet to caffe

Modify from pytorch-caffe-darknet-convert,object_detetction_tools

Modified items :

1. yolov3 output layer
2. when pooling layer stide =1 , size =2 , assign size = 1
3. upsample layer

Usage :

1. Download weights from original darknet web
2. Unmark custom_class in examples\ssd\ssd_detect.cpp
3. Remake project

> python darknet2caffe.py yolov3.cfg yolov3.weights yolov3.prototxt yolov3.caffemodel
> cd $caffe_root
> sh demo_darknet_yolov3.sh

Retrain :

You can try retrain models to approach original darknet mAP , below was my test

Network	mAP	Resolution	iters
yolov3-spp	58.7	608	100
yolov3-spp	59.0	608	200
yolov3-spp	59.8	608	1000