Refactoring CRNN and Dataset Augmentation

1. Rebuilding the CRNN model;
2. Rebuilding the EMNIST and ChineseLicelicense Plate datasets;

Model	ARCH	Model Size (MB)	EMNIST Accuracy (%)	Training Data	Testing Data
CRNN	CONV+LSTM	34	98.432	100,000	5,000
CRNN	CONV+GRU	31	98.386	100,000	5,000

Model	ARCH	Model Size (MB)	ChineseLicensePlate Accuracy (%)	Training Data	Testing Data
CRNN	CONV+LSTM	70	74.252	269,621	149,002
CRNN	CONV+GRU	58	75.649	269,621	149,002