Answers are grounded in this paper and cite their sources.
Three-dimensional (3D) documentation is an integral part of recording postmortem findings and reconstructing crime scenes. While photogrammetry is well established in forensic routine practice, deep learning-based view synthesis has introduced 3D Gaussian splatting (3DGS) as a promising technique for fast and photorealistic 3D reconstruction. However, its applicability and robustness in forensic autopsy documentation have not yet been systematically investigated. Ten forensic autopsy cases were documented using a standardized, flash-based photogrammetric workflow and reconstructed using 3DGS. Three image acquisition strategies-region-based multi-angle acquisition (RMA), fixed-angle circumferential acquisition (FCA), and continuous angular acquisition (CAA)-were applied to each case. Reconstruction quality was quantitatively assessed by comparing rendered images with reference photographs using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and learned perceptual image patch similarity (LPIPS). Reconstruction quality differed significantly between the three acquisition strategies. The CAA method consistently achieved higher PSNR and SSIM values and lower LPIPS scores than FCA and RMA. Qualitative analysis revealed method-specific artifact patterns, particularly in anatomically complex and weakly textured regions. Vignetting correction improved texture stability and reduced artifact formation. Both image acquisition strategy and photometric consistency substantially influence reconstruction quality in forensic 3DGS. Continuous angular image acquisition combined with global luminance normalization provides the most robust results. As this assessment was based on image-based quality metrics, it characterizes rendering and reconstruction quality rather than geometric or dimensional accuracy. These findings support the integration of 3DGS into forensic autopsy documentation as a complementary, primarily visual method and provide practical guidance for routine forensic implementation.
No takes yet. Share an insight, caveat, or question.
Lux et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: