Sharing a recent paper that builds upon DivPrune

[hw-ch0]

Hi,

We would like to share our recent paper accepted to ICML 2026 that builds upon your baseline.
Briefly, our framework, RESTORE, rectifies positional and attentional distortions during visual token reduction to improve accuracy while maintaining efficiency.

arXiv: https://arxiv.org/abs/2606.01711
Code: https://github.com/cvlab-yonsei/RESTORE

We appreciate your valuable contribution to the open-source research community!