This page contains the links to the articles describing the processes that Agisoft Metashape uses and is based on. Although, definitely, Metashape includes a number of optimizations, that are a part of the company know-how.
The base principles of the ground points classification method applied in Metashape Pro are described in the following scientific paper - Peter Axelsson: DEM Generation from Laser scanner data using adaptive TIN models
Classify Points (multi-class classification) - C. Becker, N. Hani, E. Rosinskaya, E. d'Angelo, C. Strecha: Classification of aerial photogrammetric 3D point clouds
Build DEM using Depth Map - Thomas Pock, Lukas Fusion, and Horst Bischof: TGV-Fusion
Metashape uses a new method for building depth maps - Patch Match (starting from 1.7):
Build Depth Map - Silvano Galliani, Katrin Lasinger, Konrad Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion
Note: in previous versions depth maps are based on the approach - Stereo Processing by Semi-Global Matching and Mutual Information, after which a dense point cloud with noise filtering is built based on the article - Real-Time Visibility-Based Fusion on Depth Maps.
Build Model from Depth Map - Nikolai Poliarnyi: Out-of-Core Surface Reconstruction via Global TGV Minimization
In the article describes the main idea of the Refine mesh tool applied in Metashape - V. H. Hiep, R. Keriven, P. Labatut, J. Pons: Towards high-resolution large-scale multi-view stereo
MicaSense image processing
Recommended MicaSense image processing technique - MicaSense RedEdge Image Processing Tutorial
Heiko Hirschmuller: Stereo Processing by Semi-Global Matching and Mutual Information (2007)