This notebook uses Shotor dataset, a synthetic dataset for word-level OCR.
This paper helped me a lot, however my architecture is not same
- https://arxiv.org/abs/1805.09441
- Pytorch Tutorial on RNNs
For word segmentation using dilation see this: - https://stackoverflow.com/a/10970473/4334320
The text of the image which I used to show the final result is a translation of this book:
- The Theory That Would Not Die, Sharon McGrayne