Training in YOLOv2 with your data

来源：互联网发布：c语言有什么用途编辑：程序博客网时间：2024/05/22 15:20

Steps for training YOLOv2 model

1. Preparing your datasets:

Put your olddatasets(Annotations,ImageSets,JPEGImages) of faster-rcnn into dirdarknet-master/scripts/VOCdevkit/VOC2007/

Edit file scripts/voc_label.py :

line7: sets=[('2007', 'trainval'), ('2007', 'test')]

line11: classes = ["mango"]

Convert data from .xml to .txt:

cd scripts/

$ python voc_label.py

Move two created files(2007_trainval.txt and2007_test.txt) to dir data/yours/ ,and rename 2007_trainval.txt to train.txt

2. Edit file data/coco.names:

xxxx

...

(List your class label names)

3. Edit file cfg/voc.data:

classes= 1

train =/home/snowflake/Downloads/darknet-master/data/voc/train.txt

valid =/home/snowflake/Downloads/darknet-master/data/voc/2007_test.txt

names = data/coco.names

backup = backup

4. Edit file cfg/tiny-yolo-voc.cfg :

line15: max_batches = 50000

line114: filters = 30 #(classes + 1 + coord)*num

line120: classes = 1 #number of ur classes

5. Train your model:

$ ./darknet detector train cfg/voc.datacfg/tiny-yolo-voc.cfg darknet.conv.weights

About 30 hours for 2w images and 5w iterations. Finalmodel will be saved in dir backup/,and you need to move them into dirbackup/yours/ to avoid being covered.

6. Detect an image:

Put your demo images into dir data/yours/

$ ./darknet detect cfg/tiny-yolo-voc.cfgbackup/yours/tiny-yolo-voc_final.weights data/yours/xxx.jpg -thresh 0.05

7. Compute the recall and avg IoU:

$ ./darknet detector recall cfg/voc.datacfg/tiny-yolo-voc.cfg backup/yours/tiny-yolo-voc_final.weights

======================================

Reference:

1.Official web: https://pjreddie.com/darknet/yolo/

2.Train ur own model:http://blog.csdn.net/sinat_30071459/article/details/53100791

3.Modify:http://lib.csdn.net/article/deeplearning/57863?knId=1726

0 0