Half-finished project of charts recognization from "Kaggle: Benetech - Making Graphs Accessible".
Get more information on Benetech - Making Graphs Accessible
Here's a brief explanation about my codes.
- I use yolo to recognize different component in the picture.
- Then manually process these information to fix x-axis and y-axis and the area of main chart.
- Depending on different type of charts, I cropped the specific part of the picture, and put it into another model.
- To process 4 types of charts, I used 4 model (manual or by yolo) to process the cropped images.
Generally, these are not successful codes. It's not robust enough to face a variety of images.
But I hope these codes can give some help or inspiration to someone in the future.