Breaking Bottlenecks: Replacing MVS Depth Map Estimation with CNNs in Archaeological Photogrammetry

Summary

This is an abstract from the session entitled "Paper / Report Submission (General Sessions)", at the 2025 annual meeting of the Society for Historical Archaeology.

As photogrammetry becomes more prevalent in archaeology and heritage preservation, computational bottlenecks increase costs and limit project scopes. Depth Map generation, a crucial yet computationally intensive step, often struggles with reflective materials. While Multi-View Stereo (MVS) is the common method for these calculations, openly available Convolutional Neural Network (CNN) depth estimation algorithms now rival traditional MVS in both speed and accuracy with reflective surfaces. This study replaces MVS depth map generation in Agisoft Metashape with various openly accessible CNNs, comparing the resulting models based on generation speed and visual acuity. To showcase the potential heritage applications of this CNN-based workflow, we model artifacts from Cabrits National Park, Dominica and an ongoing excavation on the island. We aim to evaluate the costs and benefits of implementing currently available CNNs in place of a traditional MVS-based photogrammetry workflow and discuss whether this technology may become a new standard in historical documentation and preservation.

This Resource is Part of the Following Collections

• Paper / Report Submission (General Sessions)   •
• Society for Historical Archaeology 2025

Keywords

Individual & Institutional Roles

Contact(s): Nicole Haddow