Introducing our groundbreaking graph digitizing tool, a revolutionary product currently in development. Designed to address the persistent challenges faced by academic and data-driven industries, our tool enables users to easily extract quantified data from graph images. Gone are the days of painstakingly transcribing graphs manually – simply upload an image, and our state-of-the-art computer vision techniques will do the rest.

 

The initial focus of our tool is on digitizing scatter graphs, line graphs, and bar graphs, all of which are currently in the alpha stage of development. We employ a combination of deep learning computer vision tools and classical computer vision techniques to achieve accurate results.The key component is our axes detection/recognition tool, which identifies the position of axes labels, axes title, and the overall graph title. These regions are then processed through specialized OCR models, trained to prioritize number recognition. Additionally, we utilize various computer vision techniques tailored to each graph type, such as object detection for scatter graphs, GANs for line extraction in line graphs, and linear regression of bounding boxes for isolating bars in bar graphs.

 

To ensure the utmost accuracy and robustness, our deep learning models are trained on a vast dataset of over 100,000 synthetic graphs.We have meticulously introduced a wide range of graph elements seen in each graph type during the training process, enabling our tool to handle complex scenarios with ease. We are confident that our tool outperforms humans in extracting graph data, providing results in a fraction of the time.

 

The future of our project includes the upcoming beta release of graph digitizers, starting with scatter graphs expected in late May 2023. We will continuously enhance our capabilities to handle abnormal graph variations and expand into new graph types.Join us on this transformative journey and experience the power and efficiency of our graph digitizing tool for all your data analysis needs.