CRNN Pytorch

Quick Demo

$ pip install -r requirements.txt
$ python src/predict.py -h

Everything is okay. Let's predict the demo images.

$ python src/predict.py demo/*.jpg
device: cpu
Predict: 100% [00:00<00:00,  4.89it/s]

===== result =====
demo/170_READING_62745.jpg > reading
demo/178_Showtime_70541.jpg > showtime
demo/78_Novel_52433.jpg > novel

CRNN + CTC

This is a Pytorch implementation of a Deep Neural Network for scene text recognition. It is based on the paper "An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition (2016), Baoguang Shi et al.".

Blog article with more info: https://www.ycc.idv.tw/crnn-ctc.html

Download Synth90k dataset

$ cd data
$ bash download_synth90k.sh

@InProceedings{Jaderberg14c,
  author       = "Max Jaderberg and Karen Simonyan and Andrea Vedaldi and Andrew Zisserman",
  title        = "Synthetic Data and Artificial Neural Networks for Natural Scene Text Recognition",
  booktitle    = "Workshop on Deep Learning, NIPS",
  year         = "2014",
}

@Article{Jaderberg16,
  author       = "Max Jaderberg and Karen Simonyan and Andrea Vedaldi and Andrew Zisserman",
  title        = "Reading Text in the Wild with Convolutional Neural Networks",
  journal      = "International Journal of Computer Vision",
  number       = "1",
  volume       = "116",
  pages        = "1--20",
  month        = "jan",
  year         = "2016",
}

Pretrained Model

We pretrained the RCNN model on Synth90k dataset. The weights saved at checkpoints/crnn_synth90k.pt.

Evaluate the model on the Synth90k dataset

$ python src/evaluate.py

Evaluate on 891927 Synth90k test images:

Test Loss: 0.53042

Decoded Method	Sequence Accuracy	Prediction Time
greedy	0.93873	0.44398 ms/image
beam_search (beam_size=10)	0.93892	6.9120 ms/image
prefix_beam_search (beam_size=10)	0.93900	42.598 ms/image

Train your model

You could adjust hyper-parameters in ./src/config.py.

And train crnn models,

$ python src/train.py

Acknowledgement

Please cite this repo. crnn-pytorch if you use it.

jasonminsookim/crnn-pytorch