All work
Brown Visual Computing Lab
Monocular dynamic scene decomposition (an extension of SAFF) with self-supervised vision transformer features.
- Role
- Research Assistant to Prof. James Tompkin
- Period
- Sept. 2023 – Apr. 2024
- Location
- Providence, RI
Highlights
- Investigated self-supervised vision transformer backbones (DINO v1/v2) for feature extraction, conducting ablation studies across model variants to optimize decomposition quality.
- Redesigned the post-processing pipeline with refined CRF inference and morphological filtering, improving segmentation boundary precision.