Arbitrarily applicable spatial relational responding.

For remote sensing imagery, spatial relationships are extracted by segmentation and classification. Unlike traditional vector relationships, spatial relationship reasoning is equivalent to constraint satisfaction problems in the image process. Based on 9-intersect model, the paper discuss spatial topological relationship between image objects to provide a theoretical and technical support for low levels of image segmentation, high-level analysis.

Spatial reasoning is an important component of human intelligence. We can imagine the shapes of 3D objects and reason about their spatial relations by merely looking at their three-view line drawings in 2D, with different levels of competence. Can deep networks be trained to perform spatial reasoning tasks? How can we measure their "spatial intelligence"? To answer these questions, we present the SPARE3D dataset. Based on cognitive science and psychometrics, SPARE3D contains three types of 2D-3D reasoning tasks on view consistency, camera pose, and shape generation, with increasing difficulty. We then design a method to automatically generate a large number of challenging questions with ground truth answers for each task. They are used to provide supervision for training our baseline models using state-of-the-art architectures like ResNet. Our experiments show that although convolutional networks have achieved superhuman performance in many visual learning tasks, their spatial reasoning performance on SPARE3D tasks is either lower than average human performance or even close to random guesses. We hope SPARE3D can stimulate new problem formulations and network designs for spatial reasoning to empower intelligent robots to operate effectively in the 3D world via 2D sensors. The dataset and code are available at https://ai4ce.github.io/SPARE3D.

Vision-language models (VLMs) work well in tasks ranging from image captioning to visual question answering (VQA), yet they struggle with spatial reasoning, a key skill for understanding our physical world that humans excel at.We find that spatial relations are generally rare in widely used VL datasets, with only a few being well represented while most form a long tail of underrepresented relations.This gap leaves VLMs ill-equipped to handle diverse spatial relationships.To bridge it, we construct a synthetic VQA dataset focused on spatial reasoning generated from hyperdetailed image descriptions in Localized Narratives, DOCCI, and PixMo-Cap.Our dataset consists of 455k samples containing 3.4 million QA pairs.Trained on this dataset, our Spatial-Reasoning Enhanced (SpaRE) VLMs show strong improvements on spatial reasoning benchmarks, achieving up to a 49% performance gain on the What's Up benchmark, while maintaining strong results on general tasks.Our work narrows the gap between human and VLM spatial reasoning and makes VLMs more capable in real-world tasks such as robotics and navigation.We plan to share our code and dataset in due course.

Road Network Modeling is a fundamental issue for urban computing and uses massive primitive geospatial data retrieved from Geographic Information Database. The modeling for road network is extremely complicated because of the scalable and reticular relations between the roads in the city. In this paper, we propose an approach of qualitative spatial relation and semantic web based predication for road network modeling, and define five spatial relation predicates according to the notions in point-set topology for better representing the spatial relation between roads. The roads and junctions in road network are modeled as standardized well-known text literals, and deterministic spatial realtions are calculated by spatial relation reasoning. Then, all road network elements and their relations are stored as RDF triples into LarKC, a platform for scalable semantic data processing and reasoning. In this paper, we show that the triplized road network data stored in semantic web repository is very convenient for spatial information quering and junction type calculation.Keywordsroad networkspatial relation predicatespoint-set topologyLarKC

Spatial relation between spatial objects is a very important topic for spatial reasoning, query and analysis in global GIS. However, the most popular models in current use, which are based on the 2D Euclidean space (such as raster), exist some fundamental deficiency in theory for spatial relation reasoning in the spherical digital space (SDS). In this paper, a reasoning method of spatial relations in the SDS based on manifold is approached, in which: (1) the topological definitions, properties, and descriptions of spatial objects in SDS are presented; (2) appropriate operators from set operators (i.e. intersection, difference, difference by, symmetric difference, etc.) are utilized to distinguish the spatial relations between neighboring spatial objects; and (3) the value of the Euler number of symmetric difference is used for the detailed computational results of set operations; In this method, the 8 types basic topological relations between spatial objects in SDS can be distinguished accurately and easily from simple to detailed level for different requirements.

Similarity relation is one of the spatial relations in the community of geographic information science and cartography. It is widely used in the retrieval of spatial databases, the recognition of spatial objects from images, and the description of spatial features on maps. However, little achievements have been made for it by far. In this paper, spatial similarity relation was put forward with the introduction of automated map generalization in the construction of multi-scale map databases; then the definition of spatial similarity relations was presented based on set theory, the concept of spatial similarity degree was given, and the characteristics of spatial similarity were discussed in detail, including reflexivity, symmetry, non-transitivity, self-similarity in multi-scale spaces, and scale-dependence. Finally a classification system for spatial similarity relations in multi-scale map spaces was addressed. This research may be useful to automated map generalization, spatial similarity retrieval and spatial reasoning.

Bipolarity is an important feature of spatial information, involved in the expression of preferences and constraints about spatial positioning or in pairs of opposite spatial relations such as left and right. Another important feature is imprecision which has to be taken into account to model vagueness, inherent to many spatial relations (as for instance vague expressions such as close to, to the right of), and to gain in robustness in the representations. In previous works, we have shown that fuzzy sets and fuzzy mathematical morphology are appropriate frameworks, on the one hand, to represent bipolarity and imprecision of spatial relations and, on the other hand, to combine qualitative and quantitative reasoning in description logics extended with fuzzy concrete domains. The purpose of this paper is to integrate the bipolarity feature in the latter logical framework based on bipolar and fuzzy mathematical morphology and description logics with fuzzy concrete domains. Two important issues are addressed in this paper: the modeling of the bipolarity of spatial relations at the terminological level and the integration of bipolar notions in fuzzy description logics. At last, we illustrate the potential of the proposed formalism for spatial reasoning on a simple example in brain imaging.

The Role of the Left Hemisphere in Verbal and Spatial Reasoning Tasks

Most current semantic segmentation approaches rely on deep convolutional neural networks (CNNs). However, their use of convolution operations with local receptive fields causes failures in modeling contextual spatial relations. Prior works have tried to address this issue by using graphical models or spatial propagation modules in networks. But such models often fail to capture long-range spatial relationships between entities, which leads to spatially fragmented predictions. In this work, we introduce a simple yet effective network unit, the spatial relation module, to learn and reason about global relationships between any two spatial positions, and then produce relation-enhanced feature representations. The spatial relation module is general and extensible, and can be used in a plug-and-play fashion with the existing fully convolutional network (FCN) framework. We evaluate spatial relation module-equipped networks on semantic segmentation tasks using two aerial image datasets. The networks achieve very competitive results, bringing significant improvements over baselines.

Elements for Spatial Reasoning in Construction Christian Giraud, Stephane Hanrot Pages 105-114 (1988 Proceedings of the 5th ISARC, Tokyo, Japan, ISSN 2413-5844) Abstract: According to AI techniques we see spatial reasoning, in construction, as generation and solving of goals involving a spatial representation model of buildings defining a rich taxonomy of parts and elements, and spatial relationships between these parts and elements. We define spatial representation model and spatial relationship from previous experiments in architect knowledge representation and automated surveying. The aim is to enable very abstract and short descriptions of building component assemblies, from designers at drawing- boards or from workers on sites, which can be processed and transformed in basic geometrical properties. Keywords: No keywords DOI: https://doi.org/10.22260/ISARC1988/0016 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley

A Study of Object-Location Memory Hongbin Wang (Hongbin.Wang@uth.tmc.edu) Todd R. Johnson (Todd.R.Johnson@uth.tmc.edu) Jiajie Zhang (Jiajie.Zhang@uth.tmc.edu) Yue Wang (Yue.Wang@uth.tmc.edu) School of Health Information Sciences, University of Texas Health Science Center at Houston 7000 Fannin, Suite 600, Houston, TX 77030 USA Abstract This paper aims to study the representational nature of human object-location memory. Two experiments are reported, including both performance data and eye movement data. The results show that multiple allocentric frames of reference are used to encode spatial relationships among objects and late computation in object-location memory retrieval in object- cued conditions is often inevitable. The implications on developing a general model of human spatial cognition are discussed. Introduction One important aspect of human spatial memory has to do with remembering the location of objects relative to each other. For example, you might recall that the book you read last night is on your office desk between your computer and the desk lamp. This type of memory for spatial relationships is an essential component of a more general type of memory for spatial layout and is obviously critical for many spatial tasks including locating and navigation (see Tversky, 2000, for a review). It is not clear, however, how spatial relationships among objects are encoded in memory. While it seems apparent that allocentric frames of reference (i.e., locations defined relative to external objects) rather than egocentric frames of reference (i.e., locations relative to the observer self) are often used to describe these spatial relationships, the representational and computational nature of this description is controversial (see Hunt & Waller, 1999; Klatzky, 1998). For example, are object-based spatial relationships encoded and stored directly (early computation)? Or do they have to be inferred much later at the retrieval stage (late computation)? What factors determine which representational scheme is used? In this paper we report two experiments we conducted to directly address these issues. The results show that multiple allocentric frames of reference are used to encode spatial relationships among objects and late computation in object- location memory retrieval in object-cued conditions is often inevitable. This paper consists of three major parts. In the first section, the experimental paradigm is briefly introduced. In the second section, the experiments are reported, including both performance data and eye movement data. In the final section, we briefly discuss our ongoing work of developing a computational model of the object-location memory and its implications on modeling human spatial cognition in general. The Experimental Paradigm We adopted an experimental paradigm developed by Milner and colleagues in the 1990s, which we call the Milner paradigm (e.g., Johnsrude, Owen, Crane, Milner, & Evans, 1999; Milner, Johnsrude, & Crane, 1997; Owen, Milner, Petrides, & Evans, 1996). Though their focus was on neuroimaging studies of the brain foundations of object- location memory, the Milner paradigm offers an elegant experimental design that allows a systematic evaluation of multiple schema for representing spatial relationships. In addition, the availability of neuroimaging data provides invaluable constraints on both understanding behavioral results and developing computational models (e.g., Wang, Johnson, & Zhang, 2001). There are two phases in the Milner paradigm. In the encoding phase (Figure 1A), eight drawings (objects) are individually presented on a computer screen to subjects, with each drawing accompanied by two landmarks (solid squares). Subjects are asked to remember the locations of drawings, relative to the landmarks. In the retrieval phase, subjects are presented some cues plus two identical drawings. One of the two identical drawings (target) is presented in its original location, and the other one (noise) is presented in a different location (or more accurately, it occupies the original location of another object). Subjects are required to perform a forced-choice recognition of the target, relative to the cues. Milner and colleagues originally used four retrieval conditions: 1. In the fixed-landmark condition (Figure 1C), the two landmarks were presented as cues, along with the target-noise pair. The absolute location of landmarks and objects on the screen was unchanged from their original encoding positions. 2. In the shifted-landmark condition (Figure 1D), the two landmarks were presented as cues, along with the target-noise pair. Though the spatial relationships among the landmarks/drawings remained unchanged, the absolute locations of the landmarks/drawings on the screen were shifted. 3. In the fixed-object condition (Figure 1E), two encoded drawings instead the two landmarks were

Spatial knowledge, especially survey knowledge, plays a pivotal role in people’s daily spatial tasks. However, the inherently limited visual perception and integration of spatial information often leads to inaccuracies in survey knowledge, with far-reaching consequences that affect performance, emotion, and safety in completing spatial tasks. In this study, we developed a mixed reality (MR)-based coordinate system and conducted a comparative user experiment involving 56 participants to explore whether perceiving metric and global spatial references during navigation may enhance the accuracy of user-acquired spatial knowledge. The experiment encompassed a spatial learning task where participants navigated and learned the selected indoor environment, followed by a desktop-based landmark recalling task to evaluate the accuracy of participants’ acquired survey knowledge. The results revealed the following insights: (1) the use of MR-based coordinate system significantly improved egocentric and allocentric spatial relations in participants’ survey knowledge; (2) participants’ errors in spatial relations among landmarks accumulate over distance, whereas the MR-based coordinate system helps maintain the errors at a consistent level; (3) no significant correlations were found between errors in participants’ survey knowledge and environmental familiarity, suggesting that the aiding effect of the MR-based coordinate system was independent of users’ prior knowledge of the environment; (4) participants reported increased mental demand using the MR-based coordinate system, but this did not result in diminished spatial learning performance or additional time pressure and learning efforts. These empirical results contribute to the future study of spatial knowledge acquisition and the design of spatial learning interfaces in mixed reality.

The current study aimed to develop a behavior-analytic model of analogical reasoning. In Experiments 1 and 2 subjects (adults and children) were trained and tested for the formation of four, three-member equivalence relations using a delayed matching-to-sample procedure. All subjects (Experiments 1 and 2) were exposed to tests that examined relations between equivalence and non-equivalence relations. For example, on an equivalence-equivalence relation test, the complex sample B1/C1 and the two complex comparisons B3/C3 and B3/C4 were used, and on a nonequivalence-nonequivalence relation test the complex sample B1/C2 was presented with the same two comparisons. All subjects consistently related equivalence relations to equivalence relations and nonequivalence relations to nonequivalence relations (e.g., picked B3/C3 in the presence of B1/C1 and picked B3/C4 in the presence of B1/C2). In Experiment 3, the equivalence responding, the equivalence-equivalence responding, and the nonequivalence-nonequivalence responding was successfully brought under contextual control. Finally, it was shown that the contextual cues could function successfully as comparisons, and the complex samples and comparisons could function successfully as contextual cues and samples, respectively. These data extend the equivalence paradigm and contribute to a behaviour-analytic interpretation of analogical reasoning and complex human functioning, in general.

The role of right frontal brain regions in integration of spatial relation

Visual Question Answering (VQA) is a method of answering questions in natural language based on the content of images and has been widely concerned by researchers. The existing research on the visual question answering model mainly focuses on the point of view of attention mechanism and multi-modal fusion. It only pays attention to the visual semantic features of the image in the process of image modeling, ignoring the importance of modeling the spatial relationship of visual objects. We are aiming at the existing problems of the existing VQA model research. An effective spatial relationship reasoning network model is proposed, which can combine visual object semantic reasoning and spatial relationship reasoning at the same time to realize fine-grained multi-modal reasoning and fusion. A sparse attention encoder is designed to capture contextual information effectively in the semantic reasoning module. In the spatial relationship reasoning module, the graph neural network attention mechanism is used to model the spatial relationship of visual objects, which can correctly answer complex spatial relationship reasoning questions. Finally, a practical compact self-attention (CSA) mechanism is designed to reduce the redundancy of self-attention in linear transformation and the number of model parameters and effectively improve the model's overall performance. Quantitative and qualitative experiments are conducted on the benchmark datasets of VQA 2.0 and GQA. The experimental results demonstrate that the proposed method performs favorably against the state-of-the-art approaches. Our best single model has an overall accuracy of 71.18% on the VQA 2.0 dataset and 57.59% on the GQA dataset.