Segment This Thing: Foveated Tokenization for Efficient Point-Prompted Segmentation

Tanner Schmidt, Richard Newcombe · CVPR 2025 · 2025

DOI arXiv

Status

To read

Topics

budgeted-inference, other

Feeds

P2 · NeurIPS 2027

What they did

Makes point-prompted image segmentation efficient by foveation: a crop around the prompt is tokenized with patches that become coarser with distance from the prompt, giving far fewer tokens than uniform patches.

My note

Allocates image resolution around a point prompt. Relevant to later targeted spatial queries.

BibTeX

@inproceedings{schmidt2025-segment-this-thing,
  title = {{Segment This Thing: Foveated Tokenization for Efficient Point-Prompted Segmentation}},
  author = {Tanner Schmidt and Richard Newcombe},
  booktitle = {CVPR 2025},
  year = {2025},
  doi = {10.1109/CVPR52734.2025.02740},
  eprint = {2506.11131},
  archivePrefix = {arXiv}
}