Jul 2026
GeoTrace: Geometry-Aware Trajectory Token Compression for Video Large Language Models
GeoTrace is proposed, a training-free spatiotemporal token compression framework that decomposes video evidence into exact skeleton tokens and traceable residual event tokens that offers a compact and traceable token representation for efficient and robust Video LLM inference.
Guo-Sen Xie, Mengqi Lei, Chuan Shi et al.
· arXiv.org · 0 citations