Authors: Jodh Singh
Description: For text recognition, I tested two different methods: CTC(Connectionist Temporal Classification) and attention. In OCR, using CTC for predicting labels is the de facto standard but recently, Attention has been making alot of advancements in text related tasks. So, I trained models with both Attention and CTC to compare their performance on the above dataset. I used ResNet and VGG as the feature extraction methods from the image and used 2 layer BiLSTM as the sequence modeling architecture.