Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Benchmark on Monocular Metric Depth Estimation in Wildlife Setting

Niccolò Niccoli, Lorenzo Seidenari, Ilaria Greco, Francesco Rovero

TL;DR

This work addresses the challenge of estimating metric depth from monocular wildlife imagery, where depth cues are absent and scale varies across species and environments. It presents the first wildlife-focused benchmark for monocular metric depth estimation, using a custom dataset of $93$ outdoor camera-trap images with ground-truth distances obtained from calibrated $ChARUCO$ patterns and a protocol tailored to wildlife conditions. Four state-of-the-art MDE methods—Depth Anything V2, ML Depth Pro, ZoeDepth, and Metric3D—plus a geometric baseline are evaluated, with two depth extraction strategies (median vs mean) to assess robustness; results show Depth Anything V2 achieving $MAE = $ $0.454$ m and $r = $ $0.962$, while ZoeDepth degrades significantly in outdoor settings (e.g., $MAE$ $3.087$ m). Median-depth extraction generally outperforms mean-based aggregation across methods, and the fastest yet reasonably accurate option is ZoeDepth ($0.17$s) whereas Depth Anything V2 ($0.22$s) offers the best accuracy-speed trade-off. This benchmark provides practical baselines and guidance for deploying depth estimation in wildlife monitoring, and highlights the need for wildlife-specific datasets and domain adaptation to improve real-world utility.

Abstract

Camera traps are widely used for wildlife monitoring, but extracting accurate distance measurements from monocular images remains challenging due to the lack of depth information. While monocular depth estimation (MDE) methods have advanced significantly, their performance in natural wildlife environments has not been systematically evaluated. This work introduces the first benchmark for monocular metric depth estimation in wildlife monitoring conditions. We evaluate four state-of-the-art MDE methods (Depth Anything V2, ML Depth Pro, ZoeDepth, and Metric3D) alongside a geometric baseline on 93 camera trap images with ground truth distances obtained using calibrated ChARUCO patterns. Our results demonstrate that Depth Anything V2 achieves the best overall performance with a mean absolute error of 0.454m and correlation of 0.962, while methods like ZoeDepth show significant degradation in outdoor natural environments (MAE: 3.087m). We find that median-based depth extraction consistently outperforms mean-based approaches across all deep learning methods. Additionally, we analyze computational efficiency, with ZoeDepth being fastest (0.17s per image) but least accurate, while Depth Anything V2 provides an optimal balance of accuracy and speed (0.22s per image). This benchmark establishes performance baselines for wildlife applications and provides practical guidance for implementing depth estimation in conservation monitoring systems.

Benchmark on Monocular Metric Depth Estimation in Wildlife Setting

TL;DR

This work addresses the challenge of estimating metric depth from monocular wildlife imagery, where depth cues are absent and scale varies across species and environments. It presents the first wildlife-focused benchmark for monocular metric depth estimation, using a custom dataset of outdoor camera-trap images with ground-truth distances obtained from calibrated patterns and a protocol tailored to wildlife conditions. Four state-of-the-art MDE methods—Depth Anything V2, ML Depth Pro, ZoeDepth, and Metric3D—plus a geometric baseline are evaluated, with two depth extraction strategies (median vs mean) to assess robustness; results show Depth Anything V2 achieving m and , while ZoeDepth degrades significantly in outdoor settings (e.g., m). Median-depth extraction generally outperforms mean-based aggregation across methods, and the fastest yet reasonably accurate option is ZoeDepth (s) whereas Depth Anything V2 (s) offers the best accuracy-speed trade-off. This benchmark provides practical baselines and guidance for deploying depth estimation in wildlife monitoring, and highlights the need for wildlife-specific datasets and domain adaptation to improve real-world utility.

Abstract

Camera traps are widely used for wildlife monitoring, but extracting accurate distance measurements from monocular images remains challenging due to the lack of depth information. While monocular depth estimation (MDE) methods have advanced significantly, their performance in natural wildlife environments has not been systematically evaluated. This work introduces the first benchmark for monocular metric depth estimation in wildlife monitoring conditions. We evaluate four state-of-the-art MDE methods (Depth Anything V2, ML Depth Pro, ZoeDepth, and Metric3D) alongside a geometric baseline on 93 camera trap images with ground truth distances obtained using calibrated ChARUCO patterns. Our results demonstrate that Depth Anything V2 achieves the best overall performance with a mean absolute error of 0.454m and correlation of 0.962, while methods like ZoeDepth show significant degradation in outdoor natural environments (MAE: 3.087m). We find that median-based depth extraction consistently outperforms mean-based approaches across all deep learning methods. Additionally, we analyze computational efficiency, with ZoeDepth being fastest (0.17s per image) but least accurate, while Depth Anything V2 provides an optimal balance of accuracy and speed (0.22s per image). This benchmark establishes performance baselines for wildlife applications and provides practical guidance for implementing depth estimation in conservation monitoring systems.

Paper Structure

This paper contains 21 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Monocular Depth Estimation
  4. Depth Estimation in Wildlife and Outdoor Settings
  5. Integration of Depth Estimation in Wildlife Applications
  6. Evaluated Methods
  7. Deep Learning-Based Methods
  8. Depth Anything V2
  9. ML Depth Pro
  10. ZoeDepth
  11. Metric3D
  12. Geometric Baseline Method
  13. Depth Extraction Strategies
  14. Experiments
  15. Dataset
  16. ...and 6 more sections

Figures (2)

  • Figure 1: Representative examples from our evaluation dataset showing the diversity of natural outdoor environments and imaging conditions encountered in wildlife monitoring scenarios.
  • Figure 2: Qualitative comparison of depth estimation methods on a wildlife monitoring scene. (a) Ground truth RGB image showing the outdoor environment with measurement setup. (b-e) Predicted depth maps where warmer colors (yellow/orange) indicate closer distances and cooler colors (purple/blue) represent farther distances.