Code to replicate the results of the paper "Offline continuous handwriting recognition using sequence to sequence neural networks" https://www.sciencedirect.com/science/article/pii/S0925231218301371
Software requirements:
- Tensorflow 1.4
- OpenCV 3
The experiments folder include all the code to replicate the results of the best models included in the paper. To do this, you need:
- Download the IAM and RIMES handwriting databases
- preprocess the IAM and RIMES databases using the notebooks read_IAM_database.ipynb and read_RIMES_database.ipynb
- Execute the training scripts.
The sample folder include a toy example of the model over a synthetic sequence of MNIST digits executable in google colab. - Train a seq2seq architecture over a sequence of digits generated by the MNIST dataset. - Same model can be used to the general handwriting text recognition problem
2019-05-31: Under construction. Pending to review all paths and dependencies.