Identifying shape transformations from photographs of real objects.

Filipp Schmidt,Roland W Fleming

doi:10.1371/journal.pone.0202115

Abstract

An important task of human visual cognition is to make inferences about properties of objects. One such property is an object's causal history: what happened to the object in its past (e.g., “this paper has been folded”). There is relatively little research on whether and how we make such inferences. We took photographs of objects from six different materials (‘wax’, ‘aluminum foil’, ‘gold foil’, ‘chicken wire’, ‘putty’, ‘cardboard’) transformed by one of four shape-altering transformations (‘folded’, ‘bent’, ‘crumpled’, ‘twisted’). By varying execution of transformation and viewpoint, we obtained 30 images of each material/transformation combination (720 images). We asked different groups of participants to: (1) name transformations and materials, (2) rate images with respect to the extent they belonged to each transformation or material class, and (3) classify images into the four transformation classes. Our results show that participants can infer transformations from object shape–with accuracy being modulated by object material. This inference of causal history from observed object shape shows that we can distinguish between intrinsic (material) and extrinsic (transformation) properties of the object. The separation of observed shape features by their causal origin (‘shape scission’) presumably involves both perceptual and cognitive abilities.

Highlights

An important capability of visual cognition is to make inferences about the properties of perceived objects
We plot confusion matrices to visualize the specific pattern of errors (Fig 4), with actual materials/transformations plotted against perceived materials/transformations
The repeated-measure ANOVA (rmANOVA) showed that accuracy was by trend affected by material [F(1.37,19.11) = 3.84, p = .054], with post-hoc contrasts showing that this was driven by the lower performance for wax compared to the other materials (S1 Table)

Summary

Introduction

An important capability of visual cognition is to make inferences about the properties of perceived objects. This is challenging because the proximal stimulus (e.g., retinal image) is the result of complex interactions between intrinsic and extrinsic properties. Intrinsic properties are those that tend to define what the object is; they ‘belong to’ the object and tend to persist over time; and they tend to originate from the object itself rather than external events. Orientation, lighting conditions, viewpoint, motion caused by outside events or, as we focus on here, shape-transforming insults that deform an object (e.g., crushing, denting).

Methods

Results

Conclusion