ModifiedEAST

This repo was forked from https://github.com/huoyijie/AdvancedEAST and improved by chyelang for DL class at Tsinghua.

ModifiedEAST is an algorithm used for Scene image text detect, which is primarily based on EAST:An Efficient and Accurate Scene Text Detector, and the significant improvement was also made, which make long text predictions more accurate.
If this project is helpful to you, welcome to star.

advantages

writen in keras, easy to read and run
base on EAST, an advanced text detect algorithm
easy to extend, various backbone to choose
easy to train the model
significant improvement was made, long text predictions more accurate.

project files

config file
cfg_local.py and cfg_server.py, control parameters.
you need to use --section server argument to use cfg_server.py, or cfg_local is used by default.
pre-process data
preprocess.py,resize image
label data
label.py,produce label info
define network
network.py
define loss function
losses.py
execute training
advanced_east.py and data_generator.py
predict
predict.py and nms.py
auxiliary
resnet.py: build and load pretrained weights for Resnet101, since it's not provided by Keras.
tianchi_submit.py: generate coordinates of text block into .txt for test files.
tianchi_check.py: draw text block on image for test images to verify the results.

network arch

East

ModifiedEast

setup

python 3.5.0+
tensorflow-gpu 1.4.0+ (or tensorflow 1.4.0+)
keras 2.1.4+
numpy 1.14.1+
tqdm 4.19.7+
h5py

training

prepare training data:make data root dir(icpr), copy images to root dir, and copy txts to root dir, data format details could refer to 'ICPR MTWI 2018 挑战赛二：网络图像的文本检测', Link
modify config params in cfg_local.py or config_server.py, depending on the machine you are running your codes, see default values. you can specify --section local/server for the following commands, or the default --section local will be used. if you want to specify a gpu to use, you need to modify Line 1 in cfg file.
python preprocess.py, resize image size to a square whose width is in [257, 385, 513, 641, 737], and train respectively could speed up training process.
python label.py
python advanced_east.py
python predict.py to predict. You may need to specify a valid image path in main function of predict.py.
python tianchi_submit.py to get submission file for test sets. First you need to copy icpr_mtwi_task2/image_test into cfg.data_dir.
python tianchi_check.py to verify the prediction performance. You might need to read the code to prepare some data

references

EAST:An Efficient and Accurate Scene Text Detector