Convolutional Neural Net for Identifying Traditional Tagalog Baybayin Characters
This model is a small personal project designed to bring attention to pre-colonial filipino culture. There is little practical purpose for machines to identify characters in a largely unsued/dead script, but hopefully a project like this can shed light toward the erasure of filipino and Southeast Asian culture.
The model was trained on 400+ b/w 16x16 images of various baybayin characters. All characters were written by filipinos such as myself and close friends and relatives.
Resolution and color were kept limited to ensure very few input parameters. This allowed the project to be trained on a relatively small dataset. I plan on producing more model for various underepresented languages around the world, provided I find people qualified to replicate them. I also plan a higher resolution version, once I can compile enough images to make it feasible.
This model achieves ~90% success on 100 16x16 baybayin test images.
Models such as this are extremely time intensive. They require lots of data collection and are difficult to do on a large scale. If you would like to contribute to a project like this or feel have ideas for futures projects please reach out.