CNN + RNN for Scene Text Recognition implemented using tf2 keras module. Mostly based on this and this repo. The project code is written as simply as I could. This is to make it easy to understand and debug. Also to get to know the tf2 keras module better.
Project structure:
- Model. The model definition and architecture.
- Training. Code for training.
- Test. Code for testing the pre-trained model.
- Utils. All the utility functions used.
Loss of the pre-trained model.
Some examples of pre-trained model at work.
| image | recognition |
|---|---|
 |
Rick |
 |
MORTY |
 |
pontifically |
 |
Sparkling |
 |
Bombastic |
 |
Croons |
 |
Testicles |
- pre-train the model on more data
- visualize training progress