CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport

Peng Ling, Yingda Yin, Lingting Zhu, Weikai Chen, Shengju Qian, Zeyu Hu, Xin Wang, Wenming Yang · ECCV 2026 · 2026

arXiv

Status

To read

Topics

budgeted-inference, occlusion

Feeds

P1 · ICCV 2027

What they did

Prunes visual tokens in 3D vision-language models by preserving coverage of the scene with optimal transport, instead of maximising diversity, which tends to drop representative tokens in favour of outliers.

My note

No note yet.

BibTeX

@inproceedings{ling2026-coverprune,
  title = {{CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport}},
  author = {Peng Ling and Yingda Yin and Lingting Zhu and Weikai Chen and Shengju Qian and Zeyu Hu and Xin Wang and Wenming Yang},
  booktitle = {ECCV 2026},
  year = {2026},
  eprint = {2608.13226},
  archivePrefix = {arXiv}
}