My BUE graduation project, where I intended to accurately classify images usually found on Egyptian phones (Egyptian memes, academic photos, greetings/holiday images, etc.) using a model called EPIC: Egyptian Personal Images Classifier.
In-depth details about:
- the survey done on previous state-of-the-art models used on meme-related image classification problems
- the data scraping/processing/visualizations and ML/DL models used
are shown in the project's dissertation paper.
(note: if the pdf takes too much time to load, directly download it instead)
Project Abstract:
Manually filtering Personal images (PIs) from irrelevant ones (IRIs) on one’s phone gallery can be time consuming, thus a custom dataset was created from social media platforms like Facebook and Reddit, categorized to 9 classes, then trained on variants of CNN models where the hierarchical CNNs yielded the best average f1 score of 0.871.
Project Overview:
Graduation poster:
