YuanYunshuang / useful_scripts

useful_scripts

generate_OCR_synthetic_data.py

Generation pipeline is as following:

• Randomly generate a string of numbers
• Randomly chodose a text style to write this string in an image and label each digit with a bounding box
• Do a small rotation for the image and the bounding boxes
• Perform projective transformation for the images and bounding boxes
• Add noise and blur the images

label_generator_v1.py

see generate_OCR_synthetic_data.py

label_generator_v2.py

• Randomly generate a random string
• Randomly chodose a text style to write this string in an image and label each digit with a bounding box(8dim for 4 points)
• Do a small rotation for the image and the bounding boxes
• Perform projective transformation for the images and bounding boxes
• Add noise and blur the images