Join Narek Tumanyan, YSU ML Lab Researcher and recent PhD graduate of the Weizmann Institute of Science, for a talk on:
“Understanding and Harnessing Vision Foundation Model and Architectural Priors”.

This talk explores how the rich priors embedded in visual foundation models—and the architectural priors of deep neural networks—can be exposed and leveraged to build lightweight, powerful methods for visual understanding and generation, with little to no task-specific supervision.

Key topics include:

– Disentangling structure and appearance in self-supervised Vision Transformer features for semantic appearance transfer
– Manipulating text-to-image diffusion-model features for training-free, text-driven image translation
– Using DINO-ViT features for robust long-range point tracking, including through prolonged occlusions
– Dynamic 3D reconstruction from monocular video using CNN-based motion parametrization, tracking, and depth cues

September 18
16:30
YSU AI Lab

Find more tech events happening in Armenia here