1 paper
Leyang Chen, Junyi Wu, Zhiteng Li +1
Streaming 3D reconstruction from long monocular video sequences requires maintaining a key-value (KV) cache that grows linearly with sequence length, creating a severe memory bottl…