ViewpointDepth: A New Dataset for Monocular Depth Estimation Under Viewpoint Shifts

  • 2025-02-03 16:19:35
  • Aurel Pjetri, Stefano Caprasecca, Leonardo Taccari, Matteo Simoncini, Henrique PiƱeiro Monteagudo, Wallace Walter, Douglas Coimbra de Andrade, Francesco Sambo, Andrew David Bagdanov
  • 0

Abstract

Monocular depth estimation is a critical task for autonomous driving and manyother computer vision applications. While significant progress has been made inthis field, the effects of viewpoint shifts on depth estimation models remainlargely underexplored. This paper introduces a novel dataset and evaluationmethodology to quantify the impact of different camera positions andorientations on monocular depth estimation performance. We propose a groundtruth strategy based on homography estimation and object detection, eliminatingthe need for expensive LIDAR sensors. We collect a diverse dataset of roadscenes from multiple viewpoints and use it to assess the robustness of a moderndepth estimation model to geometric shifts. After assessing the validity of ourstrategy on a public dataset, we provide valuable insights into the limitationsof current models and highlight the importance of considering viewpointvariations in real-world applications.

 

Quick Read (beta)

loading the full paper ...