ABSTRACT
Our vision is sharpest at the centre of our gaze and becomes progressively blurry into the periphery. It is widely believed that this high foveal resolution evolved at the expense of peripheral acuity. But what if this sampling scheme is actually optimal for object recognition? To test this hypothesis, we trained deep neural networks on "foveated" images mimicking how our eyes sample the visual field: objects (wherever they were in the image) were sampled at high resolution, and their surroundings were sampled with decreasing resolution away from the objects. Remarkably, networks trained with the known human peripheral blur profile yielded the best performance compared to networks trained on shallower and steeper blur profiles, and compared to baseline state-of-the-art networks trained on full resolution images. This improvement, although slight, is noteworthy since the state-of-the-art networks are already trained to saturation on these datasets. When we tested human subjects on object categorization, their accuracy deteriorated only for steeper blur profiles, which is expected since they already have peripheral blur in their eyes. Taken together, our results suggest that blurry peripheral vision may have evolved to optimize object recognition rather than merely due to wiring constraints.
Subject(s)
Pattern Recognition, Visual , Visual Perception , Fovea Centralis , Humans , Neural Networks, Computer , Visual FieldsABSTRACT
Successful engagement with the world requires the ability to predict what will happen next. Here, we investigate how the brain makes a fundamental prediction about the physical world: whether the situation in front of us is stable, and hence likely to stay the same, or unstable, and hence likely to change in the immediate future. Specifically, we ask if judgments of stability can be supported by the kinds of representations that have proven to be highly effective at visual object recognition in both machines and brains, or instead if the ability to determine the physical stability of natural scenes may require generative algorithms that simulate the physics of the world. To find out, we measured responses in both convolutional neural networks (CNNs) and the brain (using fMRI) to natural images of physically stable versus unstable scenarios. We find no evidence for generalizable representations of physical stability in either standard CNNs trained on visual object and scene classification (ImageNet), or in the human ventral visual pathway, which has long been implicated in the same process. However, in frontoparietal regions previously implicated in intuitive physical reasoning we find both scenario-invariant representations of physical stability, and higher univariate responses to unstable than stable scenes. These results demonstrate abstract representations of physical stability in the dorsal but not ventral pathway, consistent with the hypothesis that the computations underlying stability entail not just pattern classification but forward physical simulation.
Subject(s)
Brain Mapping , Brain , Brain/diagnostic imaging , Humans , Magnetic Resonance Imaging/methods , Neural Networks, Computer , Photic StimulationABSTRACT
Achieving human-like visual abilities is a holy grail for machine vision, yet precisely how insights from human vision can improve machines has remained unclear. Here, we demonstrate two key conceptual advances: First, we show that most machine vision models are systematically different from human object perception. To do so, we collected a large dataset of perceptual distances between isolated objects in humans and asked whether these perceptual data can be predicted by many common machine vision algorithms. We found that while the best algorithms explain â¼ 70 percent of the variance in the perceptual data, all the algorithms we tested make systematic errors on several types of objects. In particular, machine algorithms underestimated distances between symmetric objects compared to human perception. Second, we show that fixing these systematic biases can lead to substantial gains in classification performance. In particular, augmenting a state-of-the-art convolutional neural network with planar/reflection symmetry scores along multiple axes produced significant improvements in classification accuracy (1-10 percent) across categories. These results show that machine vision can be improved by discovering and fixing systematic differences from human vision.
Subject(s)
Algorithms , Pattern Recognition, Visual , Humans , Neural Networks, ComputerABSTRACT
Deep neural networks have revolutionized computer vision, and their object representations across layers match coarsely with visual cortical areas in the brain. However, whether these representations exhibit qualitative patterns seen in human perception or brain representations remains unresolved. Here, we recast well-known perceptual and neural phenomena in terms of distance comparisons, and ask whether they are present in feedforward deep neural networks trained for object recognition. Some phenomena were present in randomly initialized networks, such as the global advantage effect, sparseness, and relative size. Many others were present after object recognition training, such as the Thatcher effect, mirror confusion, Weber's law, relative size, multiple object normalization and correlated sparseness. Yet other phenomena were absent in trained networks, such as 3D shape processing, surface invariance, occlusion, natural parts and the global advantage. These findings indicate sufficient conditions for the emergence of these phenomena in brains and deep networks, and offer clues to the properties that could be incorporated to improve deep networks.
Subject(s)
Models, Neurological , Neural Networks, Computer , Pattern Recognition, Visual/physiology , Visual Cortex/physiology , Humans , Visual Perception/physiologyABSTRACT
The cytoarchitecture of a neuron is very important in defining morphology and ultrastructure. Although there is a wealth of information on the molecular components that make and regulate these ultrastructures, there is a dearth of understanding of how these changes occur or how they affect neurons in health and disease. Recent advances in nanoscale imaging which resolve cellular structures at the scale of tens of nanometers below the limit of diffraction enable us to understand these structures in fine detail. However, automated analysis of these images is still in its infancy. Towards this goal, attempts have been made to automate the detection and analysis of the cytoskeletal organization of microtubules. To date, evaluation of the nanoscale organization of filamentous actin (F-actin) in neuronal compartments remains challenging. Here, we present an objective paradigm for analysis which adopts supervised learning of nanoscale images of F-actin network in excitatory synapses, obtained by single molecule based super-resolution light microscopy. We have used the proposed analysis to understand the heterogeneity in the organization of F-actin in dendritic spines of primary neuronal cultures from rodents. Our results were validated using ultrastructural data obtained from platinum replica electron microscopy (PREM). The automated analysis approach was used to differentiate the heterogeneity in the nanoscale organization of F-actin in primary neuronal cultures from wild-type (WT) and a transgenic mouse model of Alzheimer's disease (APPSwe/PS1ΔE9).
Subject(s)
Actins/ultrastructure , Dendritic Spines/ultrastructure , Image Processing, Computer-Assisted/methods , Supervised Machine Learning , Animals , Female , Hippocampus/ultrastructure , In Vitro Techniques , Male , Microscopy/methods , Rats, Sprague-DawleyABSTRACT
Symmetry is a salient visual property: It is easy to detect and influences perceptual phenomena from segmentation to recognition. Yet researchers know little about its neural basis. Using recordings from single neurons in monkey IT cortex, we asked whether symmetry-being an emergent property-induces nonlinear interactions between object parts. Remarkably, we found no such deviation: Whole-object responses were always the sum of responses to the object's parts, regardless of symmetry. The only defining characteristic of symmetric objects was that they were more distinctive compared with asymmetric objects. This was a consequence of neurons preferring the same part across locations within an object. Just as mixing diverse paints produces a homogeneous overall color, adding heterogeneous parts within an asymmetric object renders it indistinct. In contrast, adding identical parts within a symmetric object renders it distinct. This distinctiveness systematically predicted human symmetry judgments, and it explains many previous observations about symmetry perception. Thus, symmetry becomes special in perception despite being driven by generic computations at the level of single neurons.
Subject(s)
Form Perception , Neurons/physiology , Pattern Recognition, Visual , Temporal Lobe/physiology , Adult , Animals , Behavior, Animal , Female , Haplorhini , Humans , Male , Young AdultABSTRACT
We perceive objects as containing a variety of attributes: local features, relations between features, internal details, and global properties. But we know little about how they combine. Here, we report a remarkably simple additive rule that governs how these diverse object attributes combine in vision. The perceived dissimilarity between two objects was accurately explained as a sum of (a) spatially tuned local contour-matching processes modulated by part decomposition; (b) differences in internal details, such as texture; (c) differences in emergent attributes, such as symmetry; and (d) differences in global properties, such as orientation or overall configuration of parts. Our results elucidate an enduring question in object vision by showing that the whole object is not a sum of its parts but a sum of its many attributes.
Subject(s)
Concept Formation , Fixation, Ocular/physiology , Signal Detection, Psychological , Visual Perception/physiology , Adult , Cues , Eye Movements/physiology , Female , Humans , Male , Orientation/physiology , Young AdultABSTRACT
Single features such as line orientation and length are known to guide visual search, but relatively little is known about how multiple features combine in search. To address this question, we investigated how search for targets differing in multiple features (intensity, length, orientation) from the distracters is related to searches for targets differing in each of the individual features. We tested race models (based on reaction times) and co-activation models (based on reciprocal of reaction times) for their ability to predict multiple feature searches. Multiple feature searches were best accounted for by a co-activation model in which feature information combined linearly (r = 0.95). This result agrees with the classic finding that these features are separable i.e., subjective dissimilarity ratings sum linearly. We then replicated the classical finding that the length and width of a rectangle are integral features-in other words, they combine nonlinearly in visual search. However, to our surprise, upon including aspect ratio as an additional feature, length and width combined linearly and this model outperformed all other models. Thus, length and width of a rectangle became separable when considered together with aspect ratio. This finding predicts that searches involving shapes with identical aspect ratio should be more difficult than searches where shapes differ in aspect ratio. We confirmed this prediction on a variety of shapes. We conclude that features in visual search co-activate linearly and demonstrate for the first time that aspect ratio is a novel feature that guides visual search.