Zero-shot learning (ZSL) aims to classify images from unseen categories, by merely utilizing seen class as the training data. Existing works on ZSL mainly leverage global features or learn regions, which, construct embeddings semantic space. However, few of them study discrimination power implied in local image regions (parts), some sense, correspond attributes, have stronger than and can thus assist transfer between seen/unseen classes. In this paper, discover (semantic) we propose...

In this paper, we aim at learning compact and discriminative linear regression models. Linear has been widely used in different problems. However, most of the existing methods exploit conventional zero-one matrix as targets, which greatly narrows flexibility model. Another major limitation these is that learned projection fails to precisely project image features target space due their weak capability. To end, present an elastic-net regularized (ENLR) framework, develop two robust models...

Semantic segmentation aims to classify every pixel of an input image. Considering the difficulty acquiring dense labels, researchers have recently been resorting weak labels alleviate annotation burden segmentation. However, existing works mainly concentrate on expanding seed pseudo within image's salient region. In this work, we propose a non-salient region object mining approach for weakly supervised semantic We introduce graph-based global reasoning unit strengthen classification...

Few-shot semantic segmentation (FSS) aims to segment unseen class objects given very few densely-annotated support images from the same class. Existing FSS methods find query object by using prototypes or directly relying on heuristic multi-scale feature fusion. However, they fail fully leverage high-order appearance relationships between features among support-query image pairs, thus leading an inaccurate localization of objects. To tackle above challenge, we propose end-to-end scale-aware...

The key challenge of zero-shot learning (ZSL) is how to infer the latent semantic knowledge between visual and attribute features on seen classes, thus achieving a desirable transfer unseen classes. Prior works either simply align global an image with its associated class vector or utilize unidirectional attention learn limited representations, which could not effectively discover intrinsic (e.g., semantics) features. To solve above dilemma, we propose Mutually Semantic Distillation Network...

Zero-shot learning (ZSL) aims to recognize novel classes by transferring semantic knowledge from seen unseen ones. Semantic is learned attribute descriptions shared between different classes, which are strong prior for localization of object representing discriminative region features enabling significant visual-semantic interaction. Although few attention-based models have attempted learn such in a single image, the transferability and visual typically neglected. In this paper, we propose...

Convolutional neural network (CNN) has achieved the state-of-the-art performance in many different visual tasks. Learned from a large-scale training data set, CNN features are much more discriminative and accurate than handcrafted features. Moreover, also transferable among domains. On other hand, traditional dictionary-based (such as BoW spatial pyramid matching) contain local structural information, which is implicitly embedded images. To further improve performance, this paper, we propose...

Compact hash code learning has been widely applied to fast similarity search owing its significantly reduced storage and highly efficient query speed. However, it is still a challenging task learn discriminative binary codes for perfectly preserving the full pairwise similarities embedded in high-dimensional real-valued features, such that promising performance can be guaranteed. To overcome this difficulty, paper, we propose novel scalable supervised asymmetric hashing (SSAH) method, which...

Writer identification is an important topic for pattern recognition and artificial intelligence. Traditional methods rely heavily on sophisticated hand-crafted features to represent the characteristics of different writers. In this paper, we propose end-to-end framework online text-independent writer by using a recurrent neural network (RNN). Specifically, handwriting data particular are represented set random hybrid strokes (RHSs). Each RHS randomly sampled short sequence representing pen...

In recent years, incomplete multi-view clustering, which studies the challenging clustering problem on missing views, has received growing research interests. Although a series of methods have been proposed to address this issue, following problems still exist: 1) Almost all existing are based shallow models, is difficult obtain discriminative common representations. 2) These generally sensitive noise or outliers since negative samples treated equally as important samples. paper, we propose...

The key challenge for few-shot semantic segmentation (FSS) is how to tailor a desirable interaction among sup-port and query features and/or their prototypes, under the episodic training scenario. Most existing FSS methods im-plement such support/query interactions by solely leveraging plain operations - e.g., cosine similarity feature concatenation segmenting objects. How-ever, these approaches usually cannot well capture intrinsic object details in images that are widely encountered FSS,...

Zero-shot learning (ZSL) tackles the unseen class recognition problem, transferring semantic knowledge from seen classes to ones. Typically, guarantee desirable transfer, a common (latent) space is adopted for associating visual and domains in ZSL. However, existing methods align by merely mitigating distribution disagreement through one-step adaptation. This strategy usually ineffective due heterogeneous nature of feature representations two domains, which intrinsically contain both...

Accurate medical image segmentation of brain tumors is necessary for the diagnosing, monitoring, and treating disease. In recent years, with gradual emergence multi-sequence magnetic resonance imaging (MRI), multi-modal MRI diagnosis has played an increasingly important role in early by providing complementary information a given lesion. Different modalities vary significantly context, as well coarse fine information. As manual identification very complicated, it usually requires lengthy...

Weakly supervised semantic segmentation with only image-level labels aims to reduce annotation costs for the task. Existing approaches generally leverage class activation maps (CAMs) locate object regions pseudo label generation. However, CAMs can discover most discriminative parts of objects, thus leading inferior pixel-level labels. To address this issue, we propose a saliency guided <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">I</b> nter-...

Learning to reject is a special kind of self-awareness (the ability know what you do not know), which an essential factor for humans become smarter. Although machine intelligence has very accurate nowadays, it lacks such and usually acts as omniscient, resulting in overconfident errors. This article presents comprehensive overview this topic from three perspectives: confidence, calibration, discrimination. Confidence important measurement the reliability model predictions. Rejection can be...

Few-shot semantic segmentation (FSS) is an important task for novel (unseen) object under the data-scarcity scenario. However, most FSS methods rely on unidirectional feature aggregation, e.g., from support prototypes to get query prediction, and high-resolution features guide low-resolution ones. This usually fails fully capture cross-resolution relationships thus leads inaccurate estimates of objects. To resolve above dilemma, we propose a cyclic memory network (CMN) directly learn read...

Labeling objects at the subordinate level typically requires expert knowledge, which is not always available from a random annotator. Accordingly, learning directly web images for fine-grained visual classification (FGVC) has attracted broad attention. However, existence of noise in huge obstacle training robust deep neural networks. In this paper, we propose novel approach to remove irrelevant samples real-world during training, and only utilize useful updating Thus, our network can...

One-shot semantic image segmentation aims to segment the object regions for novel class with only one annotated image. Recent works adopt episodic training strategy mimic expected situation at testing time. However, these existing approaches simulate test conditions too strictly during process, and thus cannot make full use of given label information. Besides, mainly focus on foreground-background target setting. They utilize binary mask labels training. In this paper, we propose leverage...

Generative adversarial networks (GANs) for (generalized) zero-shot learning (ZSL) aim to generate unseen image features when conditioned on class embeddings, each of which corresponds one unique category. Most existing works GANs ZSL by merely feeding the seen feature/class embedding (combined with random Gaussian noise) pairs into generator/discriminator a two-player minimax game. However, structure consistency distributions among real/fake features, may shift generated away from their real...

Zero-shot learning (ZSL) tackles the novel class recognition problem by transferring semantic knowledge from seen classes to unseen ones. Semantic is typically represented attribute descriptions shared between different classes, which act as strong priors for localizing object attributes that represent discriminative region features, enabling significant and sufficient visual-semantic interaction advancing ZSL. Existing attention-based models have struggled learn inferior features in a...

Zero-shot learning (ZSL) tackles the unseen class recognition problem by transferring semantic knowledge from seen classes to ones. Typically, guarantee desirable transfer, a direct embedding is adopted for associating visual and domains in ZSL. However, most existing ZSL methods focus on implicit global features or image regions space. Thus, they fail to: 1) exploit appearance relationship priors between various local single image, which corresponds information 2) learn cooperative jointly...