ABSTRACT
This paper introduces TC-StereoNet, a novel framework for temporally coherent 3D feature learning in stereo endoscopic videos, addressing flickering depth estimates and inconsistent feature representations in minimally invasive surgery.
PAPER · PDF
Loading PDF...
Key findings
TC-StereoNet integrates a Temporal Feature Alignment Module for cross-frame feature consistency.
Includes a Temporal Consistency Loss to penalize feature drift.
Employs a Memory-Augmented Feature Bank to aggregate information over time.
Validated on SCARED and Hamlyn datasets with temporal consistency metrics and clinical applicability assessment.
Limitations & open questions
Further research needed for broader clinical validation and integration into surgical workflows.