Reviews: Gated Recurrent Convolution Neural Network for OCR

Authors added gating mechanism in Recurrent Convolutional NN that allow controlling the amount of context / recursion. In other words, it is similar to improvement from RNN to GRU but for Recurrent CNNs. Positives aspects: - authors improved upon the state of the art results on very well research benchmark - the proposed model combines few proved ideas into a single model and made it work. Areas to improve: - Lines 195-196: you state that choose 10 best models per architecture that perform well on each dataset and state the results as the median of them. It is ok when you compare different iterations of you model to make sure you found good hyperparameters.