HRoPE: Homography Relative Positional Encoding

Huachen Gao^*, Yang Yang^*, Xiaoqian Liang, Xingyu Chen,
Youjia Zhang, Qi Zhang, Baorui Ma, Anpei Chen^†, Wei Yang^†

_{* Equal contribution   † Corresponding authors}

TL;DR: Scene-centric positional encoding for multi-view transformers via patch-level, plane-induced homographies.

hrope-demo-web.mp4

Coming soon.