The study examined whether the 18th-century Italian architect Giovanni Battista Piranesi used photographic references for his etchings. The researcher analyzed 20 of Piranesi's works from the Art Institute of Chicago and 102 photographs from the York Urban Database using the CLIP neural network.
When presented with original images, CLIP correctly identified all 20 etchings and 24 photographs. However, after applying a filter to turn photographs into line-based drawings, the model misclassified 16 of the 24 images as etchings, demonstrating that the presence of lines alone is not a reliable indicator of origin.
The researcher also tested for lens distortion by measuring the curvature of lines. While the program found that 96 out of 102 photographs showed no significant distortion, it also found that some genuine photos contained nearly straight lines, making this metric inconclusive for determining the use of a camera.
Additional testing with the GeoCalib neural network showed that the style of shading significantly impacts perspective estimation. The program produced inconsistent results when analyzing line-based drawings compared to photographs, rendering the method unreliable for confirming the use of photographic equipment by Piranesi.
Ultimately, the experiments could not confirm or rule out the use of photographic models. The researcher concluded that neither line straightness nor neural network classification provides definitive evidence, as artistic perspective can be constructed manually without mechanical aids.
