Real-time text detection with EAST in TFLite
Sayak Paul
Kolkata, West Bengal
- 0 Collaborators
I worked on converting the EAST model (used for text detection from natural scenes) to TFLite and published the converted model on TF Hub. ...learn more
Project status: Published/In Market
Internet of Things, Artificial Intelligence
Intel Technologies
Intel Python
Overview / Usage
Text detection from natural scenes is a pretty hard problem in computer vision and it has been studied well over the years. Because of the real-world use-cases, a text detector like this can cater to, it has become to have an optimized version of the original EAST model in TensorFlow Lite.
OCR is quite heavily used across many different industries such as finance, law enforcement, and text detection is pretty much involved in every OCR application (where the text comes in unstructured forms) that we see these days.
Methodology / Approach
I used the frozen graph of the EAST model (taken from this blog post) and converted it to an optimized form using TensorFlow Lite. I then used the model to infer on still images and live video feeds for detecting text.
Technologies Used
TensorFlow Lite