1 paper
Jiasen Lu, Liangchen Song, Mingze Xu +5
We present AToken, the first unified visual tokenizer that achieves both high-fidelity reconstruction and semantic understanding across images, videos, and 3D assets. Unlike existi…