Leveraging Synthetic Priors for Monocular Depth Estimation in Specular Surgical Environments
Ankan Aich, Yangming Lee · 2025
Read Leveraging Synthetic Priors for Monocular Depth Estimation in Specular Surgical Environments by Ankan Aich, Yangming Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-23786.txt.
Description
Ankan Aich and five co-authors improve depth estimation for surgical cameras, a capability autonomous robotic surgery depends on. A single camera must infer three-dimensional shape, but models trained outside the body fail inside it: real operations bring mirror-like reflections off wet tissue and constant deformation. Training on noisy labels from real surgery causes boundaries to collapse. The arXiv record's title governs over the catalog's shorter one.
The team instead adapts the strong synthetic priors of a foundation depth model, Depth Anything V2, using a low-rank adaptation scheme with dynamic vectors. On the public SCARED benchmark they report a new state of the art, with squared relative error cut by over seventeen percent in the most reflective conditions. They also introduce ROCAL-T 90, the first validation set from real operations, with ninety endoscopic sequences and sub-millimeter ground-truth trajectories from CT alignment.
Catalog record
| CREATOR | Ankan Aich, Yangming Lee |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 18 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-23786