TY  - RPRT
TI  - Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Composition
AU  - Seokmin Lee
AU  - Yunghee Lee
AU  - Byeonghyun Pak
AU  - Byeongju Woo
PY  - 2026
UR  - https://arxiv.org/abs/2603.13904
ID  - 2603.13904
ER  -