handsomebrothershuaili/vtd

scene text detection

Scene Text Detection and Recognition

The VTD project is on scene text detection and recognition, based on `EAST', 'CTPN' and 'CRNN'.

RoadMap

Scene Text Detection is a fast evolving field with new techniques and architectures being published frequently.
The goal of this project is facilitating the development of such techniques and applications. While constantly improving the quality of code and readme.
The main functions and models of VTD are as follows:

• Detection: EAST, CTPN and my designed network
• Recognition: CRNN
• GPU_Tracker: track your gpu usage

Usage

First of all, the data package is a preprocess lib to resize or extract the text area.
'Detection' includes these fancy models, how to use them? start with the main.py script of each model. And do not foget to change the data path for your own data.
Recognition only includes the CRNN model, and start with main.py to train your network. For details, you can read the arguments of DataGenerator.py.

Demo

Comparison

To verify the effectiveness of our proposed framework, we did a series of comparative experiments on the ICDAR2013, ICDAR2015 and ICDAR2017 data sets. These experiments mainly focus on evaluating two essential methods in our model: Channel Attention Block, and Spatial Attention Block. All the experiments are based on single model and single-scale, the scales of testing images are set as 720 for three benchmark datasets.

Author

• Shuai Li (李帅) - all work

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details