The School of Computer Science is pleased to present…
An Ensemble Deep Learning Approach for Enhanced Classification: A Case Study on Pituitary Tumors
MSc Thesis Proposal by: Sumaiya Deen Muhammad
Date: Tuesday November 14, 2023
Time: 11:00 am – 12:30 pm
Via Zoom:
Join Zoom Meeting
https://us06web.zoom.us/j/83695030670
Meeting ID: 836 9503 0670
Passcode: 182885
One tap mobile
+16699006833,,83695030670# US (San Jose)
+16892781000,,83695030670# US
Abstract: The Segment Anything Model (SAM) by Meta AI Research, trained on an extensive collection of over 1 billion masks, has gained significant attention for its exceptional ability to segment "anything" in "any scene". SAM integrates a sophisticated image encoder, prompt encoder, and lightweight mask decoder, enabling flexible prompting and rapid mask generation in segmentation tasks. This segmentation model excels in granular, component-level segmentation, enriching our understanding of pixel semantics, critical for local feature learning. On a different note, the challenge of classifying small-scale objects persists, especially in sectors like medical imaging and remote sensing where objects of interest typically represent a small fraction of the entire image.
In this study, we aim to investigate the potential applications of SAM in the classification of small objects despite its primary design as a segmentation model. We introduce an ensemble deep learning methodology that leverages SAM within our custom dataset, specifically targeting the classification of tiny objects. Through comparative analysis between segmented data (processed by SAM) and non-segmented data (original data), our findings indicate a performance improvement in favor of the segmented data, underscoring the efficacy of our proposed approach. To the best of our knowledge, our work is the first to utilize SAM’s capability in order to enhance classification of small-scale objects.
Keywords: Segment Anything Model, ensemble model, classification
Thesis Committee:
Internal Reader: Dr. Boubakeur Boufama
External Reader: Dr. Esam Abdel-Raheem
Advisor: Dr. Ziad Kobti
