Baoguang Shi

@hn18001 @rremani We have another [paper](http://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Shi_Robust_Scene_Text_CVPR_2016_paper.pdf) that has done that, but on attention-based generator rather than CRNN. But I believe the same idea would work on CRNN. In our experience,...

@Jayhello The trained model is directly usable on variable length images. If you want to use RGB images, you need to modify the model structure and retrain the model.

@NightFury13 The released model should deliver close results to the ones reported in the paper. Our best result did not reach 86% on IIIT5k. We did not include IIIT5k training...

@rremani 1.2. CRNN is only for cropped words. For whole images with much more background, a text detection method is required to detect text first. If you feed whole images...

@rremani Sorry but I am not sure if that will work -- worth a try, I think.

FYI, we have recently released another project for text detection https://github.com/MhLiao/TextBoxes.

@ll36771 I haven't tested warp_ctc yet, but I guess they implement the same algorithm. cudnn.BLSTM should also be a better choice. Please let me know if you find them workable,...

@rayush7 I guess so. Unnormalized data will probably result in divergence at the beginning of training.

Non-English characters have not been supported yet. The issue is that `tf.string_split` in r1.4 splits UTF-8 strings into bytes rather than multi-byte characters. A [PR](https://github.com/tensorflow/tensorflow/issues/18271) has been created but seems...