Abstract
Monocular depth estimation supports navigation, reconstruction, and augmented reality in endoscopic surgery, but models trained on one clinical setting often fail when deployed across hospitals, scopes, and anatomical sites. To this end, we present MEDE-TTA, a unified framework for cross-scenario monocular endoscopic depth estimation via test-time adaptation (TTA), to adapt a source-trained network for inference on unseen data without ground-truth labels. We propose Aug-Teacher to construct robust pseudo labels by averaging EMA predictions over color-jittered views, using anchor-confidence gating to switch between augmentation-averaged and standard EMA teachers. To preserve pre-trained knowledge and limiting catastrophic drift, a stochastic reset (Stoch-Reset) mechanism is proposed to restores adapted parameters toward the source snapshot after each update. Besides, a scale-invariant depth consistency loss is utilized to further stabilize adaptation. Zero-shot experiments on Hamlyn and C3VD with the baseline model respectively trained on SCARED and SimCol verify that MEDE-TTA out-performs existing TTA methods and state-of-the-art endoscopy depth estimation methods on training-free monocular depth estimation in unseen endoscopic environment. The code has been released on https://github.com/BaymaxShao/MEDE-TTA
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SafeSurg_006.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=5U4vQTCgjP
BibTex
@InProceedings{ShaLia_MEDETTA_MICCAISAT2026,
author = { Shao, Liangjing AND Cui, Beilei AND Bai, Long AND Hoque, Mobarak I. AND Ren, Hongliang},
title = { { MEDE-TTA: Cross-Scenario Monocular Endoscopy Depth Estimation via Test Time Adaptation } },
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}
}