wangpengnorman / SAR-Strong-Baseline-for-Text-Recognition

SAR: A Simple and Strong Baseline for Irregular Text Recognition

Show, Attend and Read

This is the code for the paper "Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition", Hui Li*, Peng Wang*, Chunhua Shen, Guyu Zhang (* indicates equal contribution) accepted to AAAI-19

Installation

The model is implemented in Torch, and has been tested under Ubuntu 14.04, with CUDA 8.0 and CUDNN 7.0. It depends on the following packages: torch/torch7, torch/nn, torch/nngraph, torch/image, lua-cjson, which can be easily install by "luarocks install **". CUDA-enabled GPUs are required. In addition, LMDB is required which can be installed by "apt-get install liblmdb-dev" and "pip install lmdb" in Ubuntu.

Pretrained Model

The pretrained model is localated in https://pan.baidu.com/s/1Z4a0l6UNhuWY3BDy8Z4Ctg because of the space limitation. Download it and put it into the "saved_model" folder.

Run the model

To run the model on a new image or image directory, use the script "run_model.lua".

To run the pretrained model on a provided image, use the '-input_image' flag, for example, th run_model.lua -input_image data/beach.jpg

To test the model on an entire directory of images, use the '-input_dir' flag instead: th run_model.lua -input_dir /path/to/my/image/folder

The results will be wroten into the folder vis/data.

Model training

To train the model, follow the following steps:

1. Prepare the training data, including the public available synthetic data:

   Syn90k (http://www.robots.ox.ac.uk/~vgg/data/text/)
   SynthText (http://www.robots.ox.ac.uk/~vgg/data/scenetext/)
   SynthAdd (https://pan.baidu.com/s/1uV0LtoNmcxbO-0YA7Ch4dg  (code:627x))
   

   and public available real image datasets:

   IIIT5K (http://cvit.iiit.ac.in/projects/SceneTextUnderstanding/IIIT5K.html)
   SVT (http://www.iapr-tc11.org/mediawiki/index.php/The_Street_View_Text_Dataset)
   ICDAR2013, ICDAR2015, and COCO-Text (http://rrc.cvc.uab.es/?com=introduction)
   

2. Use the script "create_dataset.py" to generate a group of "data.mdb" files which contain both synthetic and real data. The generated "data.mdb" will be saved under "DataDB" folder. To use create_dataset.py, the training images and their labels should be placed in the imagePathDir and a 'txt' labelfile separately.

3. Run the script "th main_train.lua" to train the model. The model will be saved regularly under the folder "saved_model".

Citation

Please cite the following paper if you are using the code/model in your research paper.

@InProceedings{SAR_aaai19, author = {Hui Li and Peng Wang and Chunhua Shen and Guyu Zhang}, title = {Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition}, booktitle ={AAAI Conference on Artificial Intelligence}, year = {2019} }

License

This code is only for academic purpose. For commercial purpose, please contact us ([e-mail]).