keras yolov3 h5 model file convert to darknet yolov3.weights

Introduction

This script according to https://github.com/qqwweee/keras-yolo3/blob/master/convert.py implementation yolov3 train saved h5 model convert to darknet yolov3.weights.

Check and modify files

1. voc_classes.txt: Check there is not a line break in the file end, otherwise line break will regard as one class. Just delete the line break it will be ok.

2. yolo.cfg: Modify the yolo.cfg corresponding to own train config, change the below items.

   width=960 # train image's width
   height=512 # train image's heigh

   ......
   filters=18 # 3*(5+num_classes)
   activation=linear

   [yolo]
   mask = 6,7,8
   anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
   classes=1 # classes number

   ......

   filters=18 # 3*(5+num_classes)
   activation=linear

   [yolo]
   mask = 3,4,5
   anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
   classes=1 # classes number

   ......

   filters=18 # 3*(5+num_classes)
   activation=linear

   [yolo]
   mask = 0,1,2
   anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
   classes=1 # classes number

3. open check_weight.py and modify model_path, config_path, weight_file of yourself.

model_path = "./trained_weights_final.h5"     # keras yolov3 h5 model file
config_path = 'yolov3.cfg'                   # .cfg  file path
weight_file = open('yolov3.weights', 'wb')   # save darknet yolov3 weights file path

Convert start

run python check_weight.py