Spatial tuning of face part representations within face-selective areas revealed by high-field fMRI

Regions sensitive to specific object categories as well as organized spatial patterns sensitive to different features have been found across the whole ventral temporal cortex (VTC). However, it is unclear that within each object category region, how specific feature representations are organized to support object identification. Would object features, such as object parts, be represented in fine-scale spatial tuning within object category-specific regions? Here, we used high-field 7T fMRI to examine the spatial tuning to different face parts within each face-selective region. Our results show consistent spatial tuning of face parts across individuals that within right posterior fusiform face area (pFFA) and right occipital face area (OFA), the posterior portion of each region was biased to eyes, while the anterior portion was biased to mouth and chin stimuli. Our results demonstrate that within the occipital and fusiform face processing regions, there exist systematic spatial tuning to different face parts that support further computation combining them.

Target Tracking, Navigation, and Decision-Making

This chapter explains why and how tracking of objects moving relative to an observer, and visual optic flow navigation of an observer relative to the world, are controlled by complementary cortical streams through MT--MSTv and MT+-MSTd, respectively. Target tracking uses subtractive processing of visual signals to extract an object’s bounding contours as they move relative to a background. Navigation by optic flow uses additive processing of an entire scene to derive properties such as an observer’s heading, or self-motion direction, as it moves through the scene. The chapter explains how the aperture problem for computing heading in natural scenes is solved in MT+-MSTd using a hierarchy of processing stages that is homologous to the one that solves the aperture problem for computing motion direction in MT--MSTv. Both use feedback which obeys the ART Matching Rule to select final perceptual representations and choices. Compensation for eye movements using corollary discharge, or efference copy, signals enables an accurate heading direction to be computed. Neurophysiological data about heading direction are quantitatively simulated. Log polar processing by the cortical magnification factor simplifies computation of motion direction. This space-variant processing is maximally position invariant due to the cortical choice of network parameters. How smooth pursuit occurs, and is maintained during accurate tracking, is explained. Goal approach and obstacle avoidance are explained by attractor-repeller networks. Gaussian peak shifts control steering to a goal, as well as peak shift and behavioral contrast during operant conditioning, and vector decomposition during the relative motion of object parts.

Spatial organization of face part representations within face-selective areas revealed by high-field fMRI

Regions sensitive to specific object categories as well as organized spatial patterns sensitive to different features have been found across the whole ventral temporal cortex (VTC). However, it is unclear that within each object category region, how specific feature representations are organized to support object identification. Would object features, such as object parts, be represented in fine-scale spatial organization within object category-specific regions? Here we used high-field 7T fMRI to examine the spatial organization of neural tuning to different face parts within each face-selective region. Our results show consistent spatial organization across individuals that within right posterior fusiform face area (pFFA) and right occipital face area (OFA), the posterior portion of each region was biased to eyes, while the anterior portion was biased to mouth and chin stimuli. Our results demonstrate that within the occipital and fusiform face processing regions, there exist systematic spatial organizations of neural tuning to different face parts that support further computation combining them.

Identifying solution strategies in a mental-rotation test with gender-stereotyped objects by analyzing gaze patterns

Many studies deal with solution strategies in mental-rotation tests. The approaches range from global analysis, attention to object parts, holistic and piecemeal strategy to a combined strategy. Other studies do not speak of strategies, but of holistic or piecemeal processes or even of holistic or piecemeal rotation. The methodological approach used here is to identify mental-rotation strategies via gaze patterns derived from eye-tracking data when solving chronometric mental-rotation tasks with gender-stereotyped objects. The mental-rotation test consists of 3 male-stereotyped objects (locomotive, hammer, wrench) and 3 female-stereotyped objects (pram, hand mirror, brush) rotated at eight different angles. The sample consisted of 16 women and 10 men (age: M=21.58; SD=4.21). The results of a qualitative analysis with two individual objects (wrench and brush) showed four different gaze patterns. These gaze patterns appeared with different frequency in the two objects and correlated differently with performance and response time. The results indicate either an object-oriented or an egocentric mental-rotation strategy behind the gaze patterns. In general, a new methodological approach has been developed to identify mental-rotation strategies bottom-up which can also be used for other stimulus types.

Subitizing object parts reveals a second stage of individuation

Humans can efficiently individuate a small number of objects. This subitizing ability is thought to be a consequence of limited attentional resources. However, how and what is selected during the individuation process remain outstanding questions. We investigated these in four experiments by examining if parts of objects are enumerated as efficiently as distinct objects in the presence and absence of distractor objects. We found that distractor presence reduced subitizing efficiency. Crucially, parts connected to multiple objects were enumerated less efficiently than independent objects or parts connected to a single object. These results argue against direct individuation of parts and show that objecthood plays a fundamental role in individuation. Objects are selected first and their components are selected in subsequent steps. This reveals that individuation operates sequentially over multiple levels.

Subitizing object parts reveals a second stage of individuation

Humans can efficiently individuate a small number of objects. This subitizing ability is thought to be a consequence of limited attentional resources. However, how and what is selected during the individuation process remain outstanding questions. We investigated these in three experiments by examining if parts of objects are enumerated as efficiently as distinct objects in the presence and absence of distractor objects. We found that distractor presence reduced subitizing efficiency. Crucially, parts connected to multiple objects were enumerated less efficiently than independent objects or parts connected to a single object. These results argue against direct individuation of parts and show that objecthood plays a fundamental role in individuation. Objects are selected first and their components are selected in subsequent steps. This reveals that individuation operates sequentially over multiple levels.

A Two-Branch Network for Weakly Supervised Object Localization

Weakly supervised object localization (WSOL) has attracted intense interest in computer vision for instance level annotations. As a hot research topic, a number of existing works concentrated on utilizing convolutional neural network (CNN)-based methods, which are powerful in extracting and representing features. The main challenge in CNN-based WSOL methods is to obtain features covering the entire target objects, not only the most discriminative object parts. To overcome this challenge and to improve the detection performance of feature extracting related WSOL methods, a CNN-based two-branch model was presented in this paper to locate objects using supervised learning. Our method contained two branches, including a detection branch and a self-attention branch. During the training process, the two branches interacted with each other by regarding the segmentation mask from the other branch as the pseudo ground truth labels of itself. Our model was able to focus on capturing the information of all the object parts due to the self-attention mechanism. Additionally, we embedded multi-scale detection into our two-branch method to output two-scale features. We evaluated our two-branch network on the CUB-200-2011 and VOC2007 datasets. The pointing localization, intersection over union (IoU) localization, and correct localization precision (CorLoc) results demonstrated competitive performance with other state-of-the-art methods in WSOL.

Semantic Component Association within Object Classes Based on Convex Polyhedrons

Most objects are composed of semantically distinctive parts that are more or less geometrically distinctive as well. Points on the object relevant for a certain robot operation are usually determined by various physical properties of the object, such as its dimensions or weight distribution, and by the purpose of object parts. A robot operation defined for a particular part of a representative object can be transferred and adapted to other instances of the same object class by detecting the corresponding components. In this paper, a method for semantic association of the object’s components within the object class is proposed. It is suitable for real-time robotic tasks and requires only a few previously annotated representative models. The proposed approach is based on the component association graph and a novel descriptor that describes the geometrical arrangement of the components. The method is experimentally evaluated on a challenging benchmark dataset.