Extended implementaion of CRNN network on Pytorch, for bigger image input size processing.
Original implementation: CRNN
PyTorch implementation of original crnn: Pytorch CRNN
You are able to change training parameters by editing config file number_recognition.ini
path_to_save_results - the directories for model and metrics will be created automatically in this location
model_type - this is name of model's current architecture
model_name - this is name of model with type model_type, that have been trained at dataset dataset_name
Dataset structure is showed at the image below.
Labels information is contained in labels json files: train_labels_name.json, val_labels_name.json.
Example of labels file with two images:
{file_name_1.jpg": {"number": "959205766", "type_id": "type_id_1"},
"file_name_2.jpg": {"number": "5329147", "type_id": "type_id_m"}}