23 Nov:
- Added corpus from Project Guttenberg, and randomly generated addresses.
- Added open source fonts from https://www.fontsquirrel.com/
- Added images to use as background images.
24 Nov:
- Generated images using text_renderer.
- Created tfrecords using aocr.
25 Nov:
- Started training the model.
- Tested the model.
- Noticed that the model was not predicting '@' symbol since the number of email addresses were rare in the dataset. So, created 2k more images with only email addresses and numbers.
- Created tfrecords for new images.
26 Nov:
- Merged tfrecords to make training more inclusive.
- Started training model.
- Finished training the model.
- Tested the model.
27 Nov:
- Exported the model.
- Completed writing the API.
28 Nov:
- Generated new data for training using different fonts and backgrounds.
- Changed aocr to save loss history, train configuration and training loss plot.
- Generated a small dataset of 900 images to try different model configurations and select the best one for traing with bigger dataset.
- Started training small dataset using different configurations.
29 Nov- 2 Dec Training the model.