AerialMetric: Benchmarking and Adapting UAV Monocular Metric Depth Estimation in the Real World

Zhongqiang Song¹, Guanying Chen^1✉, Yuqi Zhang^2,3, Yin Zou¹, Chuanyu Fu¹, Zhiyuan Yuan¹, Chuan Huang^4,2, Shuguang Cui^3,2, Xiaochun Cao¹

¹Sun Yat-sen University, Shenzhen Campus, ²FNii-Shenzhen, ³SSE, CUHKSZ, ⁴SIAS, USTC

ECCV 2026

Paper Code Dataset & Weight

Teaser

AerialMetric benchmark overview and qualitative metric depth comparison — **Abstract.** This paper addresses the problem of monocular metric depth estimation in aerial UAV imagery. Although recent data-driven methods have achieved remarkable progress in ground-level scenarios, models trained primarily on street-view and indoor datasets exhibit significant domain gaps when applied to aerial viewpoints.

To tackle these challenges, we introduce **AerialMetric**, a benchmark dataset designed to evaluate and facilitate the adaptation of monocular metric depth estimation under UAV aerial viewpoints. The dataset consists of four complementary subsets collected from different sources, jointly covering real-world photogrammetry data, controlled aerial acquisition settings, photorealistic synthetic scenes, and in-the-wild Internet imagery. Totally, AerialMetric provides 52K real-world and 16K synthetic image-depth pairs with reliable metric ground truth.

Based on this dataset, we conduct systematic evaluations of existing state-of-the-art models under aerial settings and investigate the impact of viewpoint, altitude, and camera parameters on metric depth prediction. In addition, by fine-tuning representative metric depth models on our dataset, we establish a comprehensive aerial benchmark and achieve state-of-the-art performance across diverse aerial imagery. Our datasets will be made publicly available.

Projected 3D Point Maps Reveal Scale Drift

We project GT, MoGe2, and MoGe2-Aerial depth into 3D, making their metric scale differences directly visible.

Input RGB Pick P1, P2, or P3.

MoGe2

MoGe2-Aerial

Data Collection Pipeline

A hybrid construction pipeline turns real captures, reconstructed scenes, and synthetic views into reliable metric depth supervision.

The AerialMetric Dataset

To address the severe domain gap between ground-level training data and aerial perspectives, we introduce AerialMetric, built through a hybrid construction pipeline. It comprises four complementary components:

Overview
Oblique
Decoupled
Synthetic
Wild

Qualitative Results

Zero-shot transfer of state-of-the-art ground-domain models (e.g., ZoeDepth, DepthPro, UniDepthV2) to aerial imagery often results in severe scale ambiguity and geometric distortion. By adapting the MoGe2 foundation model on our dataset (denoted as MoGe2-Aerial), we recover more reliable metric scale in real-world aerial scenes.

Decoupled Flight Parameter Analysis & Robustness

All parameters
FOV & Altitude
Altitude & Pitch

Parameters robustness: Baseline models exhibit severe, non-monotonic degradation regarding camera pitch and altitude, suffering catastrophic failures at strictly nadir (-90°) and highly oblique (-45°) angles. Our fine-tuned MoGe2-Aerial demonstrates exceptional stability across all evaluated pitch angles, altitudes, and FOVs.

BibTeX

@inproceedings{song2026aerialmetric,
  title     = {AerialMetric: Benchmarking and Adapting UAV Monocular Metric Depth Estimation in the Real World},
  author    = {Song, Zhongqiang and Chen, Guanying and Zhang, Yuqi and Zou, Yin and Fu, Chuanyu and Yuan, Zhiyuan and Huang, Chuan and Cui, Shuguang and Cao, Xiaochun},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}