TBrain_China Steel AI Challenge

This repository contains code for developing an AI algorithm to recognize printed and handwritten serial numbers on steel billets, aiding in tracing and quality control. The system identifies serial numbers, including digits and letters, and handles handwritten, inverted, duplicated, or blurred serial numbers for production line use.

Preparation and Process

This repository refers to CRAFT.

1. Using this repository to add some randomness while framing the input steel billet image with text areas.
2. Then use deep-text-recognition-benchmark to recognize the text in the framed steel billet image.

Code and folder description

• test.py:
  Use the weights provided by the official CRAFT for prediction.
• cut_code.py & cut_code2:
  Apply cutting methods with different randomness to the selected bounding box.
• image folder:
  Images pending processing.
• start folder:
  Images with bounding boxes awaiting cutting after running test.py.
• end folder:
  Images after being cut using randomized cutting methods.