vision transformer

Fine-Grained Visual Representations through Sparse Selection of Localized Features

Fatimah Zohra, Ph.D. Student, Computer Science

Oct 30, 14:00 - 16:00

B1 R2202

vision transformer implicit neural representation spatio-temporal modeling Vision Foundation Models visual computing

This thesis studies and develops sparse localized feature selection techniques for vision transformer (ViT) models to improve representation of fine-grained visual details across images and videos to capture fine-grained spatial and temporal structure that are often lost in global representations of current ViT foundation models.