← Back to the feed
Space & aviationResearch findings

GRAC · Improved Surgical Scene Description via Graph-Based Retrieval Augmentation

Original published 2026.11.07 03:47 KSTAdded to Blackdeck 2026.11.09 23:16 KST

What has changed?

Surgical scenes are represented as graphs capturing relationships among instruments, actions, and anatomical structures, with retrieved examples used to enhance

Key points

  1. Surgical scenes are modeled as graphs capturing relationships among instruments, actions, and anatomy.
  2. Retrieved graph-caption pairs are used to condition caption generation based on interaction comparisons.
  3. GRAC achieves higher CIDEr-D and triplet accuracy than existing baselines on benchmark datasets.

Connect it to your work

Instrument-tissue interactions between retrieved examples and input scenes are compared to condition caption generation.

What remains unproven

This research note is based on an abstract; check the original for test conditions and application scope.

Originals used for this note

Links open the original publisher in a new tab.

  1. arxivGRAC: Graph-Based Retrieval-Augmented Captioning for Surgical Scene Description