search

weinman / cnn_lstm_ctc_ocr

Tensorflow-based CNN+LSTM trained with CTC-loss for OCR
GNU General Public License v3.0
497 stars 170 forks source link

State-of-the-art #3

Closed skalinin closed 7 years ago

skalinin commented 7 years ago

Hi! Thanks for your work! Do you know which algorithm is now state-of the-art in OCR since crnn was invented two years ago? Are there some new models that are based (may be) on the crnn (convolutional and recurrent)?

novioleo commented 7 years ago

@skalinin you can read the paper of Xiang Bai who is the prof. of HUST MCLab. they did much work in word recognition~ I read some paper recently,and noticed that the accuracy not improved obviously.and some paper can't work like paper described~

skalinin commented 7 years ago

Thanks a lot!

novioleo commented 7 years ago

my pleasure

weinman commented 7 years ago

Stay tuned in November for the results of the ICDAR 2017 Robust Reading Competitions: http://rrc.cvc.uab.es

Currently, the 2015 Challenge 2.2 (word recognition) and 2.4 (end-to-end) leader boards include an attention-based CNN+RNN, as well as a CNN+RNN+CTC.