Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Immersive Multi-Elevation Multi-Seasonal Dataset for 3D Reconstruction and Visualization

Xijun Liu, Yifan Zhou, Yuxiang Guo, Rama Chellappa, Cheng Peng

TL;DR

The paper tackles the lack of holistic, real-world benchmarks for 3D reconstruction under diverse appearances and elevations. It presents a campus-scale dataset of the Johns Hopkins Homewood Campus, consisting of over 12,300 images captured across four seasons, various times of day, and elevations from ground to 120 m, using smartphones and drones. A multi-stage calibration pipeline is introduced, featuring temporal doppelgänger mitigation, ascending-view integration to bridge ground and aerial perspectives, and Procrustes-based global alignment to a campus-wide coordinate system. This resource provides a realistic benchmark for evaluating large-scale, photorealistic reconstruction methods (including NeRF-like approaches) under challenging illumination and perspective changes, enabling robust cross-view comparisons and method development.

Abstract

Significant progress has been made in photo-realistic scene reconstruction over recent years. Various disparate efforts have enabled capabilities such as multi-appearance or large-scale modeling; however, there lacks a welldesigned dataset that can evaluate the holistic progress of scene reconstruction. We introduce a collection of imagery of the Johns Hopkins Homewood Campus, acquired at different seasons, times of day, in multiple elevations, and across a large scale. We perform a multi-stage calibration process, which efficiently recover camera parameters from phone and drone cameras. This dataset can enable researchers to rigorously explore challenges in unconstrained settings, including effects of inconsistent illumination, reconstruction from large scale and from significantly different perspectives, etc.

An Immersive Multi-Elevation Multi-Seasonal Dataset for 3D Reconstruction and Visualization

TL;DR

The paper tackles the lack of holistic, real-world benchmarks for 3D reconstruction under diverse appearances and elevations. It presents a campus-scale dataset of the Johns Hopkins Homewood Campus, consisting of over 12,300 images captured across four seasons, various times of day, and elevations from ground to 120 m, using smartphones and drones. A multi-stage calibration pipeline is introduced, featuring temporal doppelgänger mitigation, ascending-view integration to bridge ground and aerial perspectives, and Procrustes-based global alignment to a campus-wide coordinate system. This resource provides a realistic benchmark for evaluating large-scale, photorealistic reconstruction methods (including NeRF-like approaches) under challenging illumination and perspective changes, enabling robust cross-view comparisons and method development.

Abstract

Significant progress has been made in photo-realistic scene reconstruction over recent years. Various disparate efforts have enabled capabilities such as multi-appearance or large-scale modeling; however, there lacks a welldesigned dataset that can evaluate the holistic progress of scene reconstruction. We introduce a collection of imagery of the Johns Hopkins Homewood Campus, acquired at different seasons, times of day, in multiple elevations, and across a large scale. We perform a multi-stage calibration process, which efficiently recover camera parameters from phone and drone cameras. This dataset can enable researchers to rigorously explore challenges in unconstrained settings, including effects of inconsistent illumination, reconstruction from large scale and from significantly different perspectives, etc.

Paper Structure

This paper contains 10 sections, 1 equation, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Scene Level Reconstruction Datasets
  4. Large-scale Reconstruction Datasets
  5. Data Collection
  6. Data Processing
  7. Large-scale Doppelgänger Mitigation
  8. Multi-elevation imagery integration
  9. Global Alignment
  10. Conclusion

Figures (3)

  • Figure 1: A visualization of the reconstructed Johns Hopkins Homewood campus based on our collected imagery over one year.
  • Figure 2: Example of the sparse reconstruction of Mason Hall with the registered camera poses from multi-elevation image sets under various appearance conditions. The displayed configurations include aerial imagery captured in Winter, Summer, and Fall. Ground-level images are taken in Summer and Fall. Two ascending image sequences are also included. These diverse viewpoints and time highlight the comprehensive data collection approach employed in our dataset.
  • Figure 3: An example of visual ambiguities, or "doppelgängers", observed in our dataset. The front and back of door of Clark Hall is similar, but should not be matched together. Naively using common feature matching algorithms leads to incorrect calibration.