Photogrammetric models should be understood primarily as constructed representations rather than neutral visual evidence, even though they are grounded in measurable physical traces.
At the level of trace, photogrammetry records real-world geometry through light, texture, and spatial correspondence across images. In this sense, it is empirically anchored and materially referential. However, the transition from trace to model involves multiple layers of mediation—camera calibration, image selection, tie-point extraction, algorithmic matching, filtering, meshing, scaling, and visualization. Each step embeds assumptions, thresholds, and exclusions that shape what becomes visible and what is suppressed.
Unlike a single photograph, a photogrammetric model is procedural and synthetic. It does not present a direct optical record of a moment in time but a statistically optimized reconstruction assembled from many viewpoints. Occlusions, reflective surfaces, shadows, and areas of low texture are selectively interpolated or omitted, producing a model that is internally coherent but not visually exhaustive. As a result, the model’s apparent completeness can obscure uncertainty and error.
Human decisions play a decisive role throughout the process: defining the area of interest, choosing capture resolution, setting reconstruction parameters, cleaning point clouds, and determining visual outputs. These decisions are often guided by project goals rather than epistemic neutrality, aligning the model with analytical, aesthetic, or narrative intentions. In this sense, photogrammetric models resemble cartographic or architectural drawings more than photographs—they are representations designed for interpretation and use.
Therefore, photogrammetric models can be considered reliable visual evidence only within clearly specified epistemic conditions: when their production protocols are transparent, uncertainties are documented, and their representational limits are acknowledged. They are best treated as situated visual arguments, not self-evident truths.
In conclusion, photogrammetry does not eliminate interpretation; it redistributes it across algorithms, workflows, and human choices. Its evidentiary value lies not in visual realism, but in the rigor with which the transformation from trace to model is made explicit and accountable.
Track2Training 