This repository contains the code necessary to train a convolutional neural network (model_training.py) on Python and a toy website to test
the model yourself (available on the web/ folder).
You can try out the page by clicking on this link.
On the python file model_training.py you can find the code necessary to train a convolutional neural network to be able to
recognize handwritten numbers in diverse positionings.
On the web/ folder a prototype web page running the trained model is available. You can try and draw on the canvas and let the program
guess what the number is.
To run the model on a Windows machine, just run run.bat. You must have python on your PATH for this script to work. It will automatically
launch a server with python and redirect your browser to it on your localhost.
Once the model is saved as an .h5 file, tensorflow.js can be used to export our model to web.
- First, install the package via pip:
pip install tensorflow.js - Create a new folder for your model:
mkdir model_folder - Finally, run the following code to export your model:
tensorflowjs_converter --input_format keras yourmodel.h5 model_folderThis will create a number of bin files (depends on your model size) and a.jsonfile containing model data. These files can be used on a web page through JavaScript to run the model, as seen on the toy web page.
- tensorflow