Development

A considerable amount of work has gone into eScript so far

but as always there is always more that can be done.

Areas of current, future and potential development are shown below.

Some development will need external help (see Collaboration for more details of how you can help).

Working on/will be working on

These facilities are either being worked on or are planned in the future.

  • Build a Tensorflow interface to give users an additional choice of NN engine
  • Build an AWS interface to give users an additional choice of NN engine
  • Develop import facilities for NN and training files from eScript website, other projects and other locations
  • Improve the code documentation
  • Improve the user documentation
  • Develop further training files
  • Improve glyph derivation-image walking facilities
  • Further improve user experience
  • Improve performance
  • User configuration of variant file generation
  • User configuration of training error rate

Remember: eScript is already a working system. The above represents a wish list for further development.

Possible in future

These facilities are possible in the future.

  • Extend the output options – currently rtf only
  • Combine output across documents
  • Output detailed results of transcription showing original document rows and derived values
  • Use of a database to store documents, results etc
  • Sharing of users' models and transcription results

Remember: eScript is already a working system. The above represents a wish list for further development.

Unlikely to happen without help

These facilities are unlikely to happen unless others join.

  • Add in a scanning interface
  • More image editing capabilities
  • Text direction other than left-right
  • Multi-language support

Remember: eScript is already a working system. The above represents a wish list for further development.

If you are interested in collaboration then please get in touch. Any reasonable offer of help from small steps to large take-over of responsibility considered. I would really like this project to reach fruition.

Tensorflow

We intend to build a Tensorflow interface but there are difficulties. Currently Tensorflow access scripts use Python as a front-end to interface to the Tensorflow engine which is written in C++.

There is now a Java interface that we can use but currently this works from the trained model stage onwards and therefore this presupposes that the training of the model has already occurred. The best we can do it seems without major development is to output the files that the training needs from eScript, the user then moves across to Python to execute the training and then back to eScript to pick up the trained model for use in subsequent transcriptions. We should be able to provide the Tensorflow scripts that will train the model against our provided training image files. If anyone has a better idea we would love to hear from them.

Amazon Machine Learning

We also plan to interface to the Amazon Web Services Machine Learning Resources. As yet this plan is in its early stages.