KAIST Visual AI Group · Graduate School of Metaverse
Chanho
Park
Can multimodal models actually use what they see — and can we verify that they do?
NeurIPS ’26 · CVPR ’26 · WACV ’26 · ICCV ’25 · CVPR ’24
- tile = one visual token of my photo (28 × 28 px)
move · click · swipe · scroll ↓
