wangzhongju / yolov3_tensorflow Goto Github PK

1. Download this file

$ git clone 'https://github.com/wangzhongju/yolov3_tensorflow.git'

2. place your data folder in '{project}/imgorg/'

e.g. there is a folder named 'test' what contained some images, now, wo can do follow this steps:

$ mv /path/to/test/folder {project}/imgorg
$ python cascade_detection.py

At last, you can see the results in '{project}/label2d'

1. format of txt_file:

img1_path bbox_1,labelname bbox_2,labelname ... img_path: the address of image, e.g. /path/for/test/test.jpg bbox: x1,y1,x2,y2 labelname: locate in '{project}/data/classes/train.names', you can rewrite it. therefore, every line of the txt_file just like: /path/to/test.jpg 23,145,342,32,1 535,232,23,56,0 there are two bboxes in picture named test.jpg, (23,145,342,32) and (535,232,23,56), one bbox labelname 1, the other labelname 0

2. check './core/config.py', make sure the path correct.
3. run command

$ python train.py

4. convert '.data-00000-of-00001' to '.pb' At the end of trianing, wo got some files located in '{project}/checkpoint/', e.g. one of them named 'yolov3_test_loss=12.342.ckpt-10.data-00000-of-00001', next, open the file '{project}/freeze_gragh.py', confirm that variable 'ckpt_file' have the same name with 'yolov3_test_loss=12.342.ckpt-10.data-00000-of-00001', run:

$ python freeze_graph.py

result: see the variable 'pb_file'