-
use Colabeler(Chinese name:精灵标注助手) to annotate your data,finally ,create JSON files.(Warning:The name of Picture can't Contain 'Blank space','Underline'.you can use 'tools/delete_x_in_batches.bat' to solve this problem.Put this file to your picture,and double-click)
-
Put all the json files(not 'output' folder.) in 'data/annotations'.
-
Put all the pictures in 'data/imgs',don't change pictre's name.
-
Complete the above steps
-
create embryo TXT file.
python 1.json2TXT.py
-
create anchors TXT
python 2.get_kmeans.py
-
Get finaly files you need
python 3.final_get_useful_data.py
-
in the 'OUT_PUT' files ,you can get the required documents: Annotations.txt、names.txt、yolo_anchors.txt
make sure you run the '3.final_get_useful_data.py'
-
'4.data_enhance_resize.py' this file will resize all pictures，scale range(0.3 0.8).Copy content from 'OUT_PUT/data_enhance_resize/data_enhance_resize.txt' file to 'data\data_configs\annotations.txt'(add ,not cover).Copy all pictures from 'OUT_PUT\data_enhance_resize\data' to 'data\data_configs\imgs',than perform 5,6,7 step
python 4.data_enhance_resize.py
-
5.data_enhance_crop.py Same method to use.Random clipping of pictures
-
6.data_enhance_Increase_noise.py Same method to use.Add noise to picture,The default noise is Gaussian.This script will consume a lot of time,Do according to your PC ability.Please read this script if you want to add other noise,it won't take long
-
7.data_enhance_Change_Bright.py,this script will change the bright of pitcure.