CAPA:
Depth Completion as Parameter-Efficient Test-Time Adaptation

Bingxin Ke1,2 Qunjie Zhou1 Jiahui Huang1 Xuanchi Ren1 Tianchang Shen1
Konrad Schindler2 Laura Leal-Taixé1 Shengyu Huang1
1NVIDIA 2ETH Zürich

TL;DR: CAPA is a framework for depth completion that adapts pre-trained depth models at test time. Given sparse geometric cues, it freezes the model backbone and uses parameter-efficient fine-tuning (like LoRA and VPT) to adapt to a specific sample (image or video). It works with any ViT-based model, and achieves state-of-the-art depth accuracy and temporal consistency.

CAPA Method Overview

Comparison with Baseline Methods

RGB+Condition
GT Depth
GT Points
Baseline Depth
Baseline Points
Ours Depth
Ours Points
00:00 00:00
Baseline Ours

Optimization Process

1/1

Applied to MoGe-2

CAPA improves both accuracy and (temporal) consistency beyond the base model.

1/1

Quantitative Results

Quantitative Results
1/1

Citation

@misc{ke2026capa,
    Author = {Bingxin Ke and Qunjie Zhou and Jiahui Huang and Xuanchi Ren and Tianchang Shen and Konrad Schindler and Laura Leal-Taixé and Shengyu Huang},
    Title = {Depth Completion as Parameter-Efficient Test-Time Adaptation},
    Year = {2026},
    Eprint = {arXiv:2602.14751},
}