Thai OCR with deep learning
- For neural network crazes
- End-to-end training - only need images of characters and a list of characters in unicode - no need to provide stop marks or character cuts
- with Accuracy
- Python 2.7.x
- Cuda Toolkit 8.0 (https://developer.nvidia.com/cuda-toolkit)
- TensorFlow r1.0 (https://www.tensorflow.org/install/)
- Run python src/train.py to train.
- Run python src/test.py to test.
- Put new fonts under the "fonts" directory.
- If fonts have different spacing diversity, the model will be hard to converge. (Need to go deeper?)
https://pvirie.wordpress.com/2016/11/23/a-challenge-to-test-neural-generative-attention/
MIT
That means free software.