Abstract
Accurate endoscopic camera relocalization in a given map is critical in minimally invasive surgery (MIS), yet remains challenging under extreme viewpoints and ambiguous anatomical geometry. Recent feed-forward 3D reconstruction methods have emerged as feasible and efficient scene representations, but they produce dense scene maps with out storing feature descriptors, rendering conventional descriptor-based localization pipelines incompatible. To enable descriptor-free relocalization on such maps, we propose EndoDefLoc, a framework that novelly formulates endoscope relocalization as a dense registration problem. Crucially, to improve localization accuracy and robustness under large viewpoint changes and in anatomical scenes with low discriminability, we extend this formulation with two task-driven innovations. First, a
rotation-invariant feature crafting (RIFC) stage, utilizing point-pair features to enhance geometry-based feature matching across extreme view point changes. Second, a geometry–appearance fusion (GAF) module integrates color cues with the aforementioned rotation-invariant geometry in a dual-branch hierarchical design to enhance feature distinctiveness. Experiments on public ex-vivo datasets demonstrate the efficacy of the proposed designs, while empirical studies further validate EndoDefLoc as a pioneering step towards descriptor-free localization for feed-forward endoscopic scene representations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COLAS_025.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=anmQ3tPI20
BibTex
@InProceedings{XuJin_EndoDefLoc_MICCAISAT2026,
author = { Xu, Jinjing AND Jiang, Chongyang AND Liu, Peng AND Li, Chenyang AND Bhasker, Nithya AND Shao, Shuwei AND Pfeiffer, Micha AND Speidel, Stefanie},
title = { { EndoDefLoc: Descriptor-Free Endoscopic Relocalization in Minimally Invasive Surgery } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS pending},
month = {pending},
page = {pending}
}