A New Dataset for Monocular Depth Estimation Under Viewpoint Shifts

2024-09-26

arXiv.org

AI Summary

This work addresses the insufficient robustness of monocular depth estimation under camera viewpoint shifts—e.g., pose variations in autonomous driving—and introduces the first benchmark explicitly designed for evaluating viewpoint-shift resilience. Methodologically, it proposes a novel LiDAR-free ground-truth generation strategy that integrates homography estimation with YOLOv8-based object detection, augmented by geometric consistency constraints. A multi-view road-scene dataset is constructed, and a systematic evaluation protocol for translational and rotational viewpoint shifts is established. Experiments reveal that state-of-the-art models suffer a 37% average increase in relative depth error under such shifts, exposing critical generalization limitations. The contribution includes a fully reproducible benchmark, a new evaluation protocol, and an open-source ground-truth generation framework—laying foundational groundwork for advancing viewpoint-robust monocular depth estimation.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements

Abstract

Monocular depth estimation is a critical task for autonomous driving and many other computer vision applications. While significant progress has been made in this field, the effects of viewpoint shifts on depth estimation models remain largely underexplored. This paper introduces a novel dataset and evaluation methodology to quantify the impact of different camera positions and orientations on monocular depth estimation performance. We propose a ground truth strategy based on homography estimation and object detection, eliminating the need for expensive lidar sensors. We collect a diverse dataset of road scenes from multiple viewpoints and use it to assess the robustness of a modern depth estimation model to geometric shifts. After assessing the validity of our strategy on a public dataset, we provide valuable insights into the limitations of current models and highlight the importance of considering viewpoint variations in real-world applications.

Problem

• Monocular Depth Estimation
• Viewpoint Change
• Model Limitations

Innovation

• Perspective Depth Dataset
• Cost-effective Solution
• Monocular Depth Estimation