This is a model for identifying the document type in an automated way (e.g. email, scientific publication, memo, etc). The model has been tested on the RVL-CDIP dataset, which is available at: http://www.cs.cmu.edu/~aharley/rvl-cdip/
To install the required libraries (tested on Ubuntu 17.11) run:
- python3 -m venv venv
- source venv/bin/activate
- pip install -r requirements.txt
Training the model from scratch
- Prepare a dataset:
- Place your document images in a folder and set the folder as the "dataset_path" in AutoDocClass.py
- Create a train, test and validation set CSV file containing "Relative Filename Path, Document Class id" pairs and "Filename, Class" as the header row*
- Train the model:
Set the model parameters in AutoDocClass.py and run the script
In progress:
- Main file callable from the terminal with tunable model parameters
- Callable model "export to disk" function
This project was developed by Roberto Calandrini during Pi School's AI programme in Fall 2017.
