Abstract
It is essential for humans to estimate the size and pose of objects in a scene to successfully function in the visually oriented world. We show that the classical problem of three-dimensional size (3D) perception in pictures can be explained by comparing human performance to the optimal geometrical solution. There exists a single viewpoint from which a photograph forms the same retinal image as the 3D scene, but retinal projection of size and shape is distorted from the real scene as the photograph is seen obliquely. We previously showed that size and shape inconstancy for the real scene results despite observers using the correct geometric back-transform, if the retinal image evokes misestimation of the viewing elevation (Akihito & Zaidi, 2020). Here, we examine how observers estimate 3D sizes in four different oblique views of pictures of objects lying on the ground in 16 different poses. Compared to estimates for the real scene, in obliquely viewed pictures, lengths of objects were seriously underestimated at fronto-parallel poses, but there were almost no changes for objects pointing at or away from the viewer. The inverse of the projection function for the length of the object gives the optimal correction function inferring correct 3D length if viewing parameters, such as viewing elevation and azimuth, are estimated correctly. Empirical correction functions had similar shapes to optimal ones, but with lower amplitude. Measurements revealed that observers systematically underestimated the viewing azimuth, similar to the fronto-parallel bias for object pose perception (Koch, Baig, & Zaidi, 2018). A model that incorporates misestimations of viewing elevation and azimuth into the geometric back-transform function provided good fits to observers’ estimates for 3D lengths from oblique views. These results add to accumulating evidence that observers use internalized geometric knowledge to perceive poses, sizes, and shapes in 3D scenes and their pictures.