Christian Bartz

You are right! I removed that parameter, thx for the hint. The `--refinement` turns on transformation parameter refinement with [inverse compositional spatial transformer networks](https://arxiv.org/abs/1612.03897). I thought this could be used...

Yes, you can use the model from `bartzi.de`, I suggest you start the training without initializing the recognition network with the pre-trained model, using the parameter `--load-localization`. The most important...

Looks alright so far, but there is an error in your code snippet: with print('{} - {}'.format(key, len(str(value)))) you are not printing the length of the array, but the length...

Yes I can give you an example: ``` python train_text_recognition.py -b 60 --blank-label 0 --char-map ../datasets/textrec/ctc_char_map.json --zoom 0.9 --area-factor 0.1 -lr 1e-4 ``` and that should be it... However, it...

Hmm, looks good to me. You do not need any box annotation (that is the whole purpose of this research work). I'm not sure why it does not work. Did...

Did you follow the installation instructions in `README.md`?

Hi, I think you are experiencing the problem with the numbers because of two reasons: 1. the network has not been trained with words that contain many numbers, so it...

Nice that it worked that well =) The problem you are experiencing right now is likely due to the fact that we use CTC-Loss for training the recognition network. This...

Yes, chainer does not have a native beam search implementation, I also found this unfortunate already^^ Funny thing is, that I was apparently wrong in my last post. The code...

Nice that it worked! You are mentioning a problem that I also faced, so far I have not been able to resolve this in a pleasing manner. One of my...