Google Summer of Code 2021 "Aztec Glyphs" Report

This Google Sumer of Code project "Visual recognition and deciphering of Aztec glyphs using Keras" is contributed by Lisardo Pérez Lugones with Red Hen Lab.

• Code Team: Tarun Nagdeve & Lisardo Pérez Lugones
• Mentors: Stephanie Wood, Jungseock Joo & Juan José Batalla Rosado

1. Introduction

My goal was to adapt a DeepLearning Mobilenet app developed by Tarun for glyph recognition, create a user Form for image upload to the Aztec hieroglyphs website -University of Oregon- or anywhere...

The main concept for the project was:

• Recreate aztecglyphs.uoregon.edu site
• Adapt a prototype working with CPU (non-GPU).
• Integrate a form or end-user webpage for upload images with browser (HTML Javascript JSON Socket-io).
• Provide images to the prototype and get the prediction using FLASK.
• The result of prediction comes back to the user via same webpage.
• New user images are stored on a filesystem subfolder.

Server specifications:

• Virtual Machine with 2 cores and 4 GB RAM.
• RedHat 7 Enterprise or CentOS 7 for a non-GPU enviroment.
• Do not reach more than 40 GB of virtual disk availability.
• Aztec hieroglypghs website is developed with Drupal 7.

At the end these were the features (development environment):

• User access via 5000 port (i.e. http://127.0.0.1:5000/)
• User can select multiple images (png, jpg, jpeg, gif or webp only)
• User request gets an ID and webpage waits for the results.
• Mobilenet prototype finds the 5 most accurated images from its own dataset.
• Print back in browser with the results per each user image.
• User images are stored on the server for potentially increase the dataset.
• User can clear results and start over.

Test and implementation:

• This project was tested in a local VM with same server specifications.
• Due to the need for access and availability of end-server admin, the final implementation in Univ. Oregon server is pending, but a functional URL is provided:

Aztec Glyphs Recognition URL

NOTE: After one hour of non-use, the URL needs some seconds to wake up. If you get an error from browser, just reload the URL.

2. Implementation

There are two main environments:

AZTECGLYPHS_LOCALHOST (for use in your local machine or VM)

The tool uses two main files:

• aztecglyphrecognition.py (Mobilenet prototype)

It's adapted from Tarun's work to Flask. While the client doesn't upload any image, wait in 'blank.html' mode, if not, upload image(s), load and analyze them, get the 5 closes images, extract the features and load the results in an array.

• aztecglyphrecognition.html

There is a fancy label for upload files button. When pressed it's hidden and 'Clear results' is shown instead. Then gets the results of the array from Mobilenet .py with a websocket and print back them in the browser. In the meanwhile a gear gif is shown waiting the predictions.

AZTECGLYPHS_SERVER_PRODUCTION (for deploy in a hosting server)

Use also two main files, the same as in localhost env but renamed and configured for production:

• app.py
• app.html

VISUAL RESULT

• Main page

• Results after upload

Updated

08-15-2021

You can have a look at my GSoC blog with all the timeline progress.

08-20-2021

Here is a functional URL for Aztec Glyphs Recognition

License

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.