Subordinate Categories Research Articles

Attentional Kernel Encoding Networks for Fine-Grained Visual Categorization

Fine-grained visual categorization aims to recognize objects from different sub-ordinate categories, which is a challenging task due to subtle visual differences between images. It is highly desired to identify discriminative regions while achieving highly non-linear compact representation for fine-grained visual categorization. However, existing methods either rely on manually defined part-based annotations to indicate the distinctive regions or operate on longitudinal vectors to capture the non-linear information, which may lose important spatial layout information. In this paper, we propose the Attentional Kernel Encoding Networks (AKEN) for fine-grained visual categorization. Specifically, the AKEN aggregates feature maps from the last convolutional layer of ConvNets to obtain a holistic feature representation. By Fourier embedding, it encodes features from both the longitudinal and transverse directions, which largely retains the spatial layout information. Moreover, we incorporate a Cascaded Attention (Cas-Attention) module to highlight local regions that distinguish among subordinate categories, enabling the AKEN to extract the most discriminative features. Working in conjunction with the attention mechanism, the proposed AKEN combines the strengths of ConvNets and kernels for non-linear feature learning, which can establish discriminative and descriptive feature representations for fine-grained image categorization. Experiments on three benchmark datasets show that the proposed AKEN delivers highly competitive performance, surpassing most existed methods and achieving state-of-the-art results.

Bi-Modal Progressive Mask Attention for Fine-Grained Recognition

Traditional fine-grained image recognition is required to distinguish different subordinate categories ( e.g. , birds species) based on the visual cues beneath raw images. Due to both small inter-class variations and large intra-class variations, it is desirable to capture the subtle differences between these sub-categories, which is crucial but challenging for fine-grained recognition. Recently, language modality aggregation has been proved as a successful technique to improve visual recognition in the experience. In this paper, we introduce an end-to-end trainable Progressive Mask Attention (PMA) model for fine-grained recognition by leveraging both visual and language modalities. Our Bi-Modal PMA model can not only stage-by-stage capture the most discriminative part in the visual modality by our mask-based fashion, but also explore the out-of-visual-domain knowledge from the language modality in an interactional alignment paradigm. Specifically, at each stage, a self-attention module is proposed to attend to the key patch from images or text descriptions. Besides, a query-relational module is designed to seize the key words/phrases of texts and further bridge the connection between two modalities. Later, the learned representations of bi-modality from multiple stages are aggregated as the final features for recognition. Our Bi-Modal PMA model only needs raw images and raw text descriptions, without requiring bounding boxes/part annotations in images or key word annotations in texts. By conducting comprehensive experiments on fine-grained benchmark datasets, we demonstrate that the proposed method achieves superior performance over the competing baselines, on either vision and language bi-modality or single visual modality.

Fine-grained visual categorization of butterfly specimens at sub-species level via a convolutional neural network with skip-connections

Fine-grained visual categorization often suffers from the challenges that the subordinate categories within an entry-level category can only be distinguished by subtler discriminations. It demands an effective algorithm to learn a multiple-perspective density distribution for precious categorization. To shape a coarse-to-fine object perception, a hierarchical convolutional neural network (CNN) denoting the skip-connections convolutional neural network (S-CCNN) was proposed, focusing on a butterfly domain at sub-species level due to the fine-grained structure of the category taxonomy. Specifically, based on the serial backbone, three skip-connections with Grating layers are established to link the earlier layers and upper layers of network, and integrated with DropConnect, exponential linear unit (ELU), and local response normalization (LRN) to alleviate over-fitting and vanishing gradient. Benefitting from the long-span skip-connections, coarse-grained context with orientation descriptions and finer-grained context with semantic representations can be both took into consideration, and they are jointly incorporated into framework. S-CCNN can achieve the cross-utilization of object-level and part-level representations, and its rationality were evidenced with both theory and practice. For effectness verification, a total of 24,836 lab-made images of butterfly specimens spanning 56 sub-species are utilized as testing samples, while 173,852 augmented images are employed for model training. S-CCNN delivers a consistent and significant boost in performance, i.e., validation accuracy achieved 94.17% and testing accuracy achieved 93.36%, which outperformed state-of-the-arts. S-CCNN can easily relish accuracy gains from skip-connections in fine-grained visual categorization of butterfly sub-species, without any bells and whistles.

Learning from Web Data Using Adversarial Discriminative Neural Networks for Fine-Grained Classification

Fine-grained classification is absorbed in recognizing the subordinate categories of one field, which need a large number of labeled images, while it is expensive to label these images. Utilizing web data has been an attractive option to meet the demands of training data for convolutional neural networks (CNNs), especially when the well-labeled data is not enough. However, directly training on such easily obtained images often leads to unsatisfactory performance due to factors such as noisy labels. This has been conventionally addressed by reducing the noise level of web data. In this paper, we take a fundamentally different view and propose an adversarial discriminative loss to advocate representation coherence between standard and web data. This is further encapsulated in a simple, scalable and end-to-end trainable multi-task learning framework. We experiment on three public datasets using large-scale web data to evaluate the effectiveness and generalizability of the proposed approach. Extensive experiments demonstrate that our approach performs favorably against the state-of-the-art methods.

Geç Çocukluk Dönemi Anadili Ediniminde Taksonomik Ulamlaştırma Becerileri

Dilin ve kültürün en temel unsurlarından olan “ulamsal sistem”, insan zihninin ulamlaştırma yetisinin bir ürünüdür. Ulamlar, ulamsal sistem içinde çeşitli ilişki biçimleri sergileyerek yer alırlar. Söz konusu ilişki biçimleri genel hatlarıyla “tematik” ve “taksonomik ilişkiler” olarak sınıflandırılabilir. Tematik ilişkiler, ulamlar arasındaki çağrışımsal, işlevsel, biçimsel görünümsel, yapısal ve benzeri ilişkileri kapsar (örn. tabak ile seramik, mutfak, çatal, çorba arasındaki ilişki). Ulamlar birbirleriyle anlamsal içerme veya kapsama ilişkisi üzerinden de ilişkiler kurarlar ki, bu tip ilişkiler, “Taşıt>Araba>Spor araba” örneğinde olduğu gibi “taksonomik ilişkiler” olarak adlandırılır. Taksonomiler bilginin bu ilişki çerçevesinde organize edildiği zihinsel yapılar olarak karşımıza çıkar. Bir taksonomi ÜST, TEMEL ve ALT düzey olmak üzere üç anlamsal düzeyden (katmandan) oluşur. ÜST düzey en genel ve kapsayıcı; TEMEL düzey orta genellikte, en kullanışlı ve en kolay anlaşılabilir; ALT düzey ise en detaylı anlam içeren düzeydir. Bu araştırmada, taksonomik organizasyon yapabilme becerisinin gelişimsel açıdan incelenmesi amacıyla 2, 4, 6. ve 8.sınıf öğrencilerine, birbiriyle bağlantılı üç test uygulanmıştır. Katılımcılardan, “Mobilya>____>____ “ gibi yarı yapılandırılmış taksonomilerde verilen boşlukları uygun ulamlarla doldurup KABUL EDİLEBİLİR DİZİLİŞler (KED) oluşturmaları beklenmiştir. Sonuçlar, KED oluşturma becerisinin yaşla doğru orantılı olarak geliştiğini göstermiştir. KED oluşturamadıkları durumlarda, katılımcıların tematik ilişkilendirmelere başvurduğu gözlemlenmiştir. Diğer yandan, katılımcıların “tamlama” ve “bileşik sözcükler” arasında ayrım yapabilme becerisi, ALT düzey ulamlara ait sonuçlar açısından ilginç görünümler sunmuştur.

Exploiting spatial relation for fine-grained image classification

Fine-Grained Image Classification (FGIC) aims to distinguish the images within a subordinate category. Recently, many FGIC methods have been proposed and huge progress has been made in the aspects of part detection and feature learning for FGIC. However, FGIC still remains a challenging task due to the large intra-class variance and small inter-class variance. To classify fine-grained images accurately, this paper proposes to exploit spatial relation to capture more discriminative details for FGIC. The proposed method contains two core modules: part selection module and representation module. The part selection module utilizes intrinsic spatial relation between object parts to select object part pairs with high discrimination power. The representation module exploits the interaction between object parts to describe the selected part pairs and construct a semantic image representation for FGIC. The proposed method is evaluated on CUB-200-2011 and FGVC-Aircraft datasets. Experimental results show that the classification accuracy of the proposed method can reach 85.5% on CUB-200-2011 and 86.9% on FGVC-Aircraft respectively, which exceed comparison methods obviously.

P300 and positive slow waves reveal the plausibility in inductive reasoning.

Category-based induction is an advanced cognitive function that is based on our category-level knowledge. Previous findings have recognized the distance effect in category-based induction: Inductive strength is affected by the hierarchical distance between the premises and conclusions. However, the neural mechanisms underlying this effect require elucidation. In the present study, we investigated the neural mechanisms of the distance effect by using EEG technology and a new experimental paradigm-category-based induction. In this paradigm, we used three hierarchical levels of categories-the subordinate category, the basic category, the superordinate category-and an irrelevant category. We further used these categories to create four types of trial that varied in the hierarchical distance between the premise and the conclusion: the subordinate-basic, the basic-superordinate, the subordinate-superordinate, and the irrelevant-superordinate trials. In each trial, participants judged the probability that the conclusion category had the same property as the premise category. Our behavioral results revealed that people responded more slowly in the irrelevant-superordinate trials than in the basic-superordinate and the subordinate-basic trials. Our ERP results showed that the irrelevant-superordinate trials elicited smaller P300 (250-500ms) amplitudes than did the subordinate-basic and the basic-superordinate trials. In addition, the subordinate-superordinate trials elicited smaller P300 and PSW (700-998ms) amplitudes than did the subordinate-basic and the basic-superordinate combinations. These findings indicate that the amplitudes of P300 and PSW may reflect the distance effect in inductive reasoning: The further the premise-conclusion hierarchical distance, the lower the inductive strength, and thus the smaller the P300 and PSW amplitudes.

The psychometric properties of the communicative competence scale for individuals with Aphasia using speech-generating devices

ABSTRACTBackground: Persons with aphasia (PWA) use augmentative and alternative communication (AAC) systems to access and create symbol-based messages. The purpose of this study was to (a) investigate the psychometric properties of the newly developed Communicative Competence Scale (CCS), and (b) to determine if any factors, or subcategories, exist within each scale of the scale. The CCS aims to assess the competence of PWA using AAC.Aims: Respondents were 55 undergraduate and graduate speech-language pathology students. The students were recruited to test the statistical properties of the CCS and Conversational Skills Rating Scale (CSRS). Prior to completing both instruments, respondents watched two videos of a PWA communicating with a speech-generating device. The messages were displayed using a grid display in one video and a scene display in the other.Main contribution: Concurrent validity was established between the CCS and the CSRS, and a moderate correlation was found. Reliability Coefficients revealed that the CCS and the CSRS had acceptable levels of internal consistency. Factor analysis revealed factor loadings or subordinate categories within the CCS and CSRS that were consistent with previous research.Conclusions: The CCS was found to be an internally consistent, reliable, and valid measure of perceived communicative competence of PWA using aided AAC. Clinical implications are discussed.

A Unified Matrix-Based Convolutional Neural Network for Fine-Grained Image Classification of Wheat Leaf Diseases

Fine-grained image classification methods often suffer from the challenge that the subordinate categories within an entry-level category can only be distinguished by subtle differences. Crop disease classification is affected by various visual interferences, including uneven illumination, dew, and equipment jitter. It demands an effective algorithm to accurately discriminate one category from the others. Thus, the representational ability of algorithm needs to be strengthened to learn a robust domain-specific discrimination through an effective way. To address this challenge, a unified convolutional neural network (CNN) denoting the matrix-based convolutional neural network (M-bCNN) was proposed. Its hallmark is the convolutional kernel matrix, whose convolutional layers are arranged parallelly in the form of a matrix, and integrated with DropConnect, exponential linear unit, local response normalization, and so on to defeat over-fitting and vanishing gradient. With a tolerable addition of parameters, it can effectively increase the data streams, neurons, and link channels of the model compared with the commonly used plain networks. Therefore, it will create more non-linear mappings and will enhance the representational ability with a tolerable growth of parameters. The images of winter wheat leaf diseases were utilized as experimental samples for their strong similarities among sub-categories. A total of 16652 images containing eight categories were collected from Shandong Province, China, and were augmented into 83260 images. The M-bCNN delivered significant improvements and achieved an average validation accuracy of 96.5% and a testing accuracy of 90.1%; this outperformed AlexNet and VGG-16. The M-bCNN demonstrated accuracy gains with a convolutional kernel matrix in fine-grained image classification.

Probability Fusion Decision Framework of Multiple Deep Neural Networks for Fine-Grained Visual Classification

Fine-grained visual classification tasks often suffer from that the subordinate categories within a basic-level category have low inter-class discrepancy and high intra-class variances, which is still challenging research for traditional deep neural networks (DNNs). However, different models extract local parts’ features in isolation and neglect the inherent correlations and distribution in high-dimensional space, which limit the single model to achieve better accuracy. In this paper, we propose a novel probability fusion decision framework (named as PFDM-Net) for fine-grained visual classification. More specifically, it first employs data-augmented tricks to enlarge the dataset and pretrain the basic VGG19 and ResNet networks on high-quality images datasets to learn common and domain knowledge simultaneously while fine-tuning with professional skill. Next, refined multiple DNNs with transfer learning are applied to design a multi-stream feature extractor, which utilizes the mixture-granularity information to exploit high-dimensionality features for distinguishing interclass discrepancy and tolerating intra-class variances. Finally, a probability fusion module equipped with gating network and probability fusion layer is developed to fuse different components model with Gaussian distribution as a unified probability representation for the ultimate fine-grained recognition. The input of this module is the various features of multi-models and the output is the fused classification probability. The end-to-end implementation of our framework contain an inner loop about the EM algorithm within an outer loop with the gradient back-propagation optimization of the whole network. Experimental results demonstrate the outperforming performance of PFDM-Net with higher classification accuracy on different fine-grained datasets compared with the state-of-the-arts methods. More discussions are provided to indicate the potential applications in combination with other work.

The Capacity for Correlated Semantic Memories in the Cortex.

A statistical analysis of semantic memory should reflect the complex, multifactorial structure of the relations among its items. Still, a dominant paradigm in the study of semantic memory has been the idea that the mental representation of concepts is structured along a simple branching tree spanned by superordinate and subordinate categories. We propose a generative model of item representation with correlations that overcomes the limitations of a tree structure. The items are generated through “factors” that represent semantic features or real-world attributes. The correlation between items has its source in the extent to which items share such factors and the strength of such factors: if many factors are balanced, correlations are overall low; whereas if a few factors dominate, they become strong. Our model allows for correlations that are neither trivial nor hierarchical, but may reproduce the general spectrum of correlations present in a dataset of nouns. We find that such correlations reduce the storage capacity of a Potts network to a limited extent, so that the number of concepts that can be stored and retrieved in a large, human-scale cortical network may still be of order 107, as originally estimated without correlations. When this storage capacity is exceeded, however, retrieval fails completely only for balanced factors; above a critical degree of imbalance, a phase transition leads to a regime where the network still extracts considerable information about the cued item, even if not recovering its detailed representation: partial categorization seems to emerge spontaneously as a consequence of the dominance of particular factors, rather than being imposed ad hoc. We argue this to be a relevant model of semantic memory resilience in Tulving’s remember/know paradigms.

Data Synthesize for Semantic Fine-Grained Prediction of Lung Nodule in CT Images

Background/AimSemantic terms of lung nodules such as lobulation, spiculation, malignancy, etc. can be commonly found in radiology reports to describe the semantic characteristics for diagnostic reference. The prevalence of malignancy among pulmonary nodules ranges between 0.2% and 50%. The main purpose of fine-grained prediction is to differentiate subordinate categories of the same semantic base classes and to provide an imaging biomarker with richer quantitative assessments of nodules for better prevalence estimation and diagnostic decision.ChallengesSince some subtypes or stages of one disease are rare but important, significant data imbalance might present and render the prediction problem very arduous. Data imbalance could bias the learning frameworks and is essential in the popularization of data-hungry deep learning paradigms. However, this problem was not elaborated in previous works of lung nodules.MethodsWe explored the cutting-edge Wasserstein Generative Adversarial Networks (WGAN) to address the data imbalance issue with over-sampling on the minority classes and tested this method on the public Lung Imaging Database Consortium (LIDC) dataset. The WGAN can estimate the underlying distribution of a minority class to synthesize more plausible and helpful samples for the prediction model.ResultsThe experiment shows that the proposed method improves F-measure by an average of 9.53% for the smallest minority classes compared to conventional method. It suggests that the WGAN technique can synthesize helpful samples for minority classes to assist the training of the Convolutional Neural Network (CNN) model and to boost fine-grained prediction performance.ConclusionMore synthesized samples obtained by considering the underlying distribution of minority classes will benifit the prediction. Future work will explore the relation among semantic terms for the data synthesis to simplify the oversampling process as well as to boost the synthetic quality even further.

Visual statistical learning at basic and subordinate category levels in real-world images.

Visual statistical learning (VSL) has been proposed as a powerful mechanism underlying the striking ability of human observers to handle complex visual environments. Previous studies have shown that VSL can occur when statistical information is embedded at multiple levels of abstraction, such as at semantically different category levels. In the present study, we further examined whether statistical regularities at a basic category level (e.g., a regular sequence of a bird, then a car, and then a dog) could influence the ability to extract statistical regularities at the subordinate level (e.g., a regular sequence of a parrot, then a sports car, and then an Eskimo dog). In the familiarization phase, participants were exposed to a stream of real-world images whose semantic categories had temporal regularities. Importantly, the temporal regularities existed at both the basic and subordinate levels, or the regularities existed at only the subordinate level, depending on the experimental condition. After completing the familiarization, participants performed a surprise two-alternative forced choice (2AFC) task for a familiarity judgment between two triplets in which the temporal regularities were either preserved or not preserved. Our results showed that the existence of statistical regularities at the basic level did not influence VSL at the subordinate level. The subsequent experiments showed these results consistently even when the basic-level categories had to be explicitly recognized and when the stimuli were not easily categorized at their subordinate level. Our results suggest that VSL is constrained to learn a particular level of patterns when patterns are presented across multiple levels.

Cascaded one-vs-rest detection network for fine-grained recognition without part annotations

Fine-grained recognition is a challenging task due to small intra-category variances. Most of the top-performing fine-grained recognition methods leverage parts of objects for better performance. Therefore, part annotations which are extremely computationally expensive are required. In this paper, we propose a novel cascaded deep CNN detection framework for fine-grained recognition which is trained to detect a whole object without considering parts. Nevertheless, most of the current top-performing detection networks use N + 1 class (N object categories plus background) softmax loss. The background category with much more training samples dominates the feature learning progress where the features are not suitable for object categorisation with fewer samples. To address this issue, we here introduce two strategies: 1) We leverage a cascaded structure to eliminate the background. 2) We introduce a novel one-vs-rest loss function to capture more minute variances from different subordinate categories. Experiments show that our proposed recognition framework achieves comparable performance against the state-of-the-art, part-free, fine-grained recognition methods on the CUB-200-2011 Bird dataset. Meanwhile, our method outperforms most of the existing part annotation based methods and does not need part annotations at the training stage whilst being free from any annotations at the test stage.

Visualisation of Animals by Children: How Do They See Birds?

Children learn to recognise animals from their earliest years through actual sightings in their own observations of their world, but also through second-hand representations in various forms of media. Young learners begin with a template specimen to which they refer when they see another animal that resembles it, naming the animal accordingly. Gradually, they learn to distinguish members of the subordinate category – bird in the case of the present paper – into subcategories. Accessing their mental model through drawings is one means of discerning their interpretation of both phyla and species. If children of increasing ages are studied, a rationale for the understanding of a such concepts may be forthcoming. The present study investigated children from 6 years to 14 years though interviews, as well as through the drawings on which the paper focuses. As children mature, they observe more and more detailsabout the birds that they see, thus increasing their knowledge not from school but from their own observations outside school.

The conceptual structure of positive emotions in Japanese university and graduate students

This study examined the semantic organization of positive emotions among university and graduate students in Japan. A total of 100 participants individually sorted 132 positive-emotion words into categories using flash cards. Hierarchical cluster analysis of the sorted terms yielded a comprehensive map of the concept. We found, at the highest level of abstraction, two large distinctions: “other-oriented positive emotions” and “self-oriented positive emotions.” The former included (a) aijo (love), (b) kantan (wonder), and (c) ikei/sonkei (awe/respect), while the latter included (d) nagomi/yorokobi (peacefulness/joy), (e) ikigomi (enthusiasm), and (f) hokori (pride) at the basic level. Further, these basic categories contained 11 subordinate categories. These findings suggest that in semantic organization, positive emotions are generally classified according to the self–other dimension and that 11 subordinate categories have different features.

Mask-CNN: Localizing parts and selecting descriptors for fine-grained bird species categorization

Fine-grained image recognition is a challenging computer vision problem, due to the small inter-class variations caused by highly similar subordinate categories, and the large intra-class variations in poses, scales and rotations. In this paper, we prove that selecting useful deep descriptors contributes well to fine-grained image recognition. Specifically, a novel Mask-CNN model without the fully connected layers is proposed. Based on the part annotations, the proposed model consists of a fully convolutional network to both locate the discriminative parts (e.g., head and torso), and more importantly generate weighted object/part masks for selecting useful and meaningful convolutional descriptors. After that, a three-stream Mask-CNN model is built for aggregating the selected object- and part-level descriptors simultaneously. Thanks to discarding the parameter redundant fully connected layers, our Mask-CNN has a small feature dimensionality and efficient inference speed by comparing with other fine-grained approaches. Furthermore, we obtain a new state-of-the-art accuracy on two challenging fine-grained bird species categorization datasets, which validates the effectiveness of both the descriptor selection scheme and the proposed Mask-CNN model.

Memory-Based Specification of Verbal Features for Classifying Animals into Super-Ordinate and Sub-Ordinate Categories

Accumulating evidence suggests that category representations are based on features. Distinguishing features are considered to define categories, because of all-or-none responses for objects in different categories; however it is unclear how distinguishing features actually classify objects at various category levels. The present study included 75 animals within three classes (mammal, bird, fish), along with 195 verbal features. Healthy adults participated in memory-based feature-animal matching verification tests. Analyses included a hierarchical clustering analysis, Support Vector Machine, and Independent component analysis to specify features effective for classifications. Quantitative and qualitative comparisons for significant features were conducted between super-ordinate and sub-ordinate levels. The number of significant features was larger for super-ordinate than sub-ordinate levels. Qualitatively, the proportion of biological features was larger than cultural/affective features in both the levels, while the proportion of affective features increased at the sub-ordinate level. To summarize, the two types of features differentially function to establish category representations.

The Role of Spatial Frequency Information in Face Classification by Race.

It was found that face classification by race is more quickly for other-race than own-race faces (other-race classification advantage, ORCA). Controlling the spatial frequencies of face images, the current study investigated the perceptual processing differences based on spatial frequencies between own-race and other-race faces that might account for the ORCA. Regardless of the races of the observers, the own-race faces were classified faster and more accurately for broad-band faces than for both lower and higher spatial frequency (SF) faces, whereas, although other-race faces were classified less accurately for higher SF than for either broad-band or lower SF faces, there was no difference between broad-band and lower SF conditions of other-race faces. Although it was not evident for higher SF condition, the ORCA was more evident for lower SF than that for broad-band faces. The present data indicate that global/configural information is needed for subordinate race categorization of faces and that an important source of ORCA is application of global/configural computations by default while categorizing an own-race face but not while categorizing an other-race face.

Gender, Migration, and Exclusionary Citizenship Regimes: Conceptualizing Transnational Abandonment of Wives as a Form of Violence Against Women.

Based on life history narratives of 57 women in India and interviews with 21 practitioners, we document the neglect, abuse, and instrumental deprivation of women's rights through the process of transnational abandonment. While gendered local sociocultural milieus and economic norms contribute to these harms, they are crucially enabled and sustained by transnational formal-legal frameworks. Widening the explanatory lens for understanding domestic violence beyond the family and community, we argue that in a globalized world, (inter)state policies serve to construct these women as a subordinate category of citizens-"disposable women"-who can be abused and abandoned with impunity.