A Grey parrot, Griffin (Psittacus erithacus), previously taught English labels for various colors and shapes with respect to three-dimensional (3D) stimuli, was tested on his ability to transfer to very different two-dimensional (2D) images consisting of modal and amodal completion stimuli. For modal completion (aka subjective contours), Kanizsa figures were constructed using black 'pac-men' to...

We explore the problem of content-aware image fill with convolutional neural networks. Given an image that is partially masked, our goal is to generate realistic-looking content to fill the masked parts of the image. This task is also sometimes referred to as image completion or image inpainting. We experiment with several different network architectures for the problem, and we observe our most...

A new class of dynamic volume completion is introduced, where image elements (e.g., occluding semi-ellipses placed at the edge of an object) can link across a gap between two or more objects, leading to the perception of illusory volumes that deform as those image elements are set into relative motion. These new demonstrations provide further evidence that volume completion is not dictated sole...

Amodal completion refers to the phenomenological finding of perceiving partly occluded objects as continuing uninterrupted behind an occluder. The outlying problem is how the visual system processes such non-local stimuli because the known processes of early vision are spatially restricted operations which segregate local differences in the visual image, and little is known about their interact...

In a study of visual after-images of patients with homonymous hemianopia one would anticipate an after-image only in that part of the field which is perceived by the patient during fixation. Fuchs (1920, 1921), however, found that such patients saw more in the after-image than in the stimulus field or the perceived field of vision. Thus Fuchs noted that a figure only partly seen on fixation (th...

Hyperspectral images (HSIs) have higher spectral resolution than RGB and are used in various tasks. However, HSIs prone to degradation due noise generated during imaging, making it difficult obtain non-degraded images. Additionally, supervised learning, which relies on pairs of degraded images, is often challenging apply HSI restoration because the high cost imaging need prepare large amounts d...