Search

As a fundamental and challenging task in bridging language and vision domains, Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality, and its key challenge is to measure the semantic similarity across different modalities. Although significant progress has been achieved, existing approaches typically suffer from two major limitations: (1) It hurts the accuracy of the representation by directly exploiting the bottom-up attention based region-level features where each region is equally treated. (2) It limits the scale of negative sample pairs by employing the mini-batch based end-to-end training mechanism. To address these limitations, we propose a Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR. Specifically, we delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules. One learns the global representation via the self-attention algorithm, noted as Self-Guided Enhancement (SGE) module. The other module benefits from the pre-trained CLIP module, which provides a novel scheme to exploit and transfer the knowledge from an off-the-shelf model, noted as CLIP-Guided Enhancement (CGE) module. Moreover, we incorporate the training mechanism of MoCo into ITR, in which two dynamic queues are employed to enrich and enlarge the scale of negative sample pairs. Meanwhile, a Unified Training Objective (UTO) is developed to learn from mini-batch based and dynamic queue based samples. Extensive experiments on the benchmark MSCOCO and Flickr30K datasets demonstrate the superiority of both retrieval accuracy and inference efficiency. For instance, compared with the existing best method NAAF, the metric R@1 of our USER on the MSCOCO 5K Testing set is improved by 5% and 2.4% on caption retrieval and image retrieval without any external knowledge or pre-trained model while enjoying over 60 times faster inference speed. Our source code will be released at https://github.com/zhangy0822/USER. [ABSTRACT FROM AUTHOR]

Few-shot learning (FSL) for human–object interaction (HOI) aims at recognizing various relationships between human actions and surrounding objects only from a few samples. It is a challenging vision task, in which the diversity and interactivity of human actions result in great difficulty to learn an adaptive classifier to catch ambiguous interclass information. Therefore, traditional FSL methods usually perform unsatisfactorily in complex HOI scenes. To this end, we propose dynamic graph-in-graph networks (DGIG-Net), a novel graph prototypes framework to learn a dynamic metric space by embedding a visual subgraph to a task-oriented cross-modal graph for few-shot HOI. Specifically, we first build a knowledge reconstruction graph to learn latent representations for HOI categories by reconstructing the relationship among visual features, which generates visual representations under the category distribution of every task. Then, a dynamic relation graph integrates both reconstructible visual nodes and dynamic task-oriented semantic information to explore a graph metric space for HOI class prototypes, which applies the discriminative information from the similarities among actions or objects. We validate DGIG-Net on multiple benchmark datasets, on which it largely outperforms existing FSL approaches and achieves state-of-the-art results. [ABSTRACT FROM AUTHOR]

Scene graph generation (SGG) is built on top of detected objects to predict object pairwise visual relations for describing the image content abstraction. Existing works have revealed that if the links between objects are given as prior knowledge, the performance of SGG is significantly improved. Inspired by this observation, in this article, we propose a relation regularized network (R2-Net), which can predict whether there is a relationship between two objects and encode this relation into object feature refinement and better SGG. Specifically, we first construct an affinity matrix among detected objects to represent the probability of a relationship between two objects. Graph convolution networks (GCNs) over this relation affinity matrix are then used as object encoders, producing relation-regularized representations of objects. With these relation-regularized features, our R2-Net can effectively refine object labels and generate scene graphs. Extensive experiments are conducted on the visual genome dataset for three SGG tasks (i.e., predicate classification, scene graph classification, and scene graph detection), demonstrating the effectiveness of our proposed method. Ablation studies also verify the key roles of our proposed components in performance improvement. [ABSTRACT FROM AUTHOR]

Remote sensing image captioning (RSIC), which aims at generating a well-formed sentence for a remote sensing image, has attracted more attention in recent years. The general framework for RSIC is the encoder–decoder architecture containing two submodels of encoder and decoder. Although the significant performance is obtained, the encoder–decoder architecture is a black-box model with a lack of explainability. To overcome this drawback, in this article, we propose a new explainable word–sentence framework for RSIC. The proposed word–sentence framework consists of two parts: word extractor and sentence generator, where the former extracts the valuable words in the given remote sensing image, while the latter organizes these words into a well-formed sentence. The proposed framework decomposes RSIC into a word classification task and a word sorting task, which is more in line with human intuitive understanding. On the basis of the word–sentence framework, some ablation experiments are conducted on the three public RSIC data sets of Sydney-captions, UCM-captions, and RSICD to explore the specific and effective network structures. In order to evaluate the proposed word–sentence framework objectively, we further conduct some comparative experiments on these three data sets and achieve comparable results in comparison with the encoder–decoder-based methods. [ABSTRACT FROM AUTHOR]

Extreme instance imbalance among categories and combinatorial explosion make the recognition of Human-Object Interaction (HOI) a challenging task. Few studies have addressed both challenges directly. Motivated by the success of few-shot learning that learns a robust model from a few instances, we formulate HOI as a few-shot task in a meta-learning framework to alleviate the above challenges. Due to the fact that the intrinsical characteristic of HOI is diverse and interactive, we propose a Semantic-guided Attentive Prototypes Network (SAPNet) framework to learn a semantic-guided metric space where HOI recognition can be performed by computing distances to attentive prototypes of each class. Specifically, the model generates attentive prototypes guided by the category names of actions and objects, which highlight the commonalities of images from the same class in HOI. In addition, we design two alternative prototypes calculation methods, i.e., Prototypes Shift (PS) approach and Hallucinatory Graph Prototypes (HGP) approach, which explore to learn a suitable category prototypes representations in HOI. Finally, in order to realize the task of few-shot HOI, we reorganize 2 HOI benchmark datasets with 2 split strategies, i.e., HICO-NN, TUHOI-NN, HICO-NF, and TUHOI-NF. Extensive experimental results on these datasets have demonstrated the effectiveness of our proposed SAPNet approach. [ABSTRACT FROM AUTHOR]

Recently, remote sensing image captioning (RSIC) has drawn an increasing attention. In this field, the encoder–decoder-based methods have become the mainstream due to their excellent performance. In the encoder–decoder framework, the convolutional neural network (CNN) is used to encode a remote sensing image into a semantic feature vector, and a sequence model such as long short-term memory (LSTM) is subsequently adopted to generate a content-related caption based on the feature vector. During the traditional training stage, the probability of the target word at each time step is forcibly optimized to 1 by the cross entropy (CE) loss. However, because of the variability and ambiguity of possible image captions, the target word could be replaced by other words like its synonyms, and therefore, such an optimization strategy would result in the overfitting of the network. In this article, we explore the overfitting phenomenon in the RSIC caused by CE loss and correspondingly propose a new truncation cross entropy (TCE) loss, aiming to alleviate the overfitting problem. In order to verify the effectiveness of the proposed approach, extensive comparison experiments are performed on three public RSIC data sets, including UCM-captions, Sydney-captions, and RSICD. The state-of-the-art result of Sydney-captions and RSICD and the competitive results of UCM-captions achieved by TCE loss demonstrate that the proposed method is beneficial to RSIC. [ABSTRACT FROM AUTHOR]

Vision-to-language tasks aim to integrate computer vision and natural language processing together, which has attracted the attention of many researchers. For typical approaches, they encode image into feature representations and decode it into natural language sentences. While they neglect high-level semantic concepts and subtle relationships between image regions and natural language elements. To make full use of these information, this paper attempt to exploit the text-guided attention and semantic-guided attention (SA) to find the more correlated spatial information and reduce the semantic gap between vision and language. Our method includes two-level attention networks. One is the text-guided attention network which is used to select the text-related regions. The other is SA network which is used to highlight the concept-related regions and the region-related concepts. At last, all these information are incorporated to generate captions or answers. Practically, image captioning and visual question answering experiments have been carried out, and the experimental results have shown the excellent performance of the proposed approach. [ABSTRACT FROM AUTHOR]

Convolutional Neural Network (CNN) has shown their advantages in salient object detection. CNN can generate great saliency maps because it can obtain high-level semantic information. And the semantic information is usually achieved by stacking multiple convolutional layers and pooling layers. However, multiple pooling operations will reduce the size of the feature map and easily blur the boundary of the salient object. Therefore, such operations are not beneficial to generate great saliency results. To alleviate this issue, we propose a novel edge information-guided hierarchical feature fusion network (HFFNet). Our network fuses features hierarchically and retains accurate semantic information and clear edge information effectively. Specifically, we extract image features from different levels of VGG. Then, we fuse the features hierarchically to generate high-level semantic information and low-level edge information. In order to retain better information at different levels, we adopt a one-to-one hierarchical supervision strategy to supervise the generation of low-level information and high-level information respectively. Finally, we use low-level edge information to guide the saliency map generation, and the edge guidance fusion is able to identify saliency regions effectively. The proposed HFFNet has been extensively evaluated on five traditional benchmark datasets. The experimental results demonstrate that the proposed model is fairly effective in salient object detection compared with 10 state-of-the-art models under different evaluation indicators, and it is superior to most of the comparison models. [ABSTRACT FROM AUTHOR]

Smoke density estimation from a single image is a totally new but highly ill-posed problem. To solve the problem, we stack several convolutional encoder-decoder structures together to propose a wave-shaped neural network, termed W-Net. Stacking encoder-decoders directly increases the network depth, leading to the enlargement of receptive fields for encoding more semantic information. To maximize the degrees of feature re-usage, we copy and resize the outputs of encoding layers to corresponding decoding layers, and then concatenate them to implement short-cut connections for improving spatial accuracy. The crests and troughs of W-Net are special structures containing abundant localization and semantic information, so we also use short-cut connections between these structures and decoding layers. Estimated smoke density is useful in many applications, such as smoke segmentation, smoke detection, disaster simulation. Experimental results show that our method outperforms existing methods on both smoke density estimation and segmentation. It also achieves satisfying results in visual detection of auto exhausts. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of supervised video summarization by formulating it as a sequence-to-sequence learning problem, where the input is a sequence of original video frames, and the output is a keyshot sequence. Our key idea is to learn a deep summarization network with attention mechanism to mimic the way of selecting the keyshots of human. To this end, we propose a novel video summarization framework named attentive encoder–decoder networks for video summarization (AVS), in which the encoder uses a bidirectional long short-term memory (BiLSTM) to encode the contextual information among the input video frames. As for the decoder, two attention-based LSTM networks are explored by using additive and multiplicative objective functions, respectively. Extensive experiments are conducted on two video summarization benchmark datasets, i.e., SumMe and TVSum. The results demonstrate the superiority of the proposed AVS-based approaches against the state-of-the-art approaches, with remarkable improvements on both datasets. [ABSTRACT FROM AUTHOR]

Video captioning has been attracting broad research attention in the multimedia community. However, most existing approaches heavily rely on static visual information or partially capture the local temporal knowledge (e.g., within 16 frames), thus hardly describing motions accurately from a global view. In this paper, we propose a novel video captioning framework, which integrates bidirectional long-short term memory (BiLSTM) and a soft attention mechanism to generate better global representations for videos as well as enhance the recognition of lasting motions in videos. To generate video captions, we exploit another long-short term memory as a decoder to fully explore global contextual information. The benefits of our proposed method are two fold: 1) the BiLSTM structure comprehensively preserves global temporal and visual information and 2) the soft attention mechanism enables a language decoder to recognize and focus on principle targets from the complex content. We verify the effectiveness of our proposed video captioning framework on two widely used benchmarks, that is, microsoft video description corpus and MSR-video to text, and the experimental results demonstrate the superiority of the proposed approach compared to several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Sufficient training examples are the fundamental requirement for most of the learning tasks. However, collecting well-labelled training examples is costly. Inspired by Zero-shot Learning (ZSL) that can make use of visual attributes or natural language semantics as an intermediate level clue to associate low-level features with high-level classes, in a novel extension of this idea, we aim to synthesise training data for novel classes using only semantic attributes. Despite the simplicity of this idea, there are several challenges. First, how to prevent the synthesised data from over-fitting to training classes? Second, how to guarantee the synthesised data is discriminative for ZSL tasks? Third, we observe that only a few dimensions of the learnt features gain high variances whereas most of the remaining dimensions are not informative. Thus, the question is how to make the concentrated information diffuse to most of the dimensions of synthesised data. To address the above issues, we propose a novel embedding algorithm named Unseen Visual Data Synthesis (UVDS) that projects semantic features to the high-dimensional visual feature space. Two main techniques are introduced in our proposed algorithm. (1) We introduce a latent embedding space which aims to reconcile the structural difference between the visual and semantic spaces, meanwhile preserve the local structure. (2) We propose a novel Diffusion Regularisation (DR) that explicitly forces the variances to diffuse over most dimensions of the synthesised data. By an orthogonal rotation (more precisely, an orthogonal transformation), DR can remove the redundant correlated attributes and further alleviate the over-fitting problem. On four benchmark datasets, we demonstrate the benefit of using synthesised unseen data for zero-shot learning. Extensive experimental results suggest that our proposed approach significantly outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Scene recognition has been widely studied to understand visual information from the level of objects and their relationships. Toward scene recognition, many methods have been proposed. They, however, encounter difficulty to improve the accuracy, mainly due to two limitations: 1) lack of analysis of intrinsic relationships across different scales, say, the initial input and its down-sampled versions and 2) existence of redundant features. This paper develops a semi-supervised learning mechanism to reduce the above two limitations. To address the first limitation, we propose a multitask model to integrate scene images of different resolutions. For the second limitation, we build a model of sparse feature selection-based manifold regularization (SFSMR) to select the optimal information and preserve the underlying manifold structure of data. SFSMR coordinates the advantages of sparse feature selection and manifold regulation. Finally, we link the multitask model and SFSMR, and propose the semi-supervised learning method to reduce the two limitations. Experimental results report the improvements of the accuracy in scene recognition. [ABSTRACT FROM PUBLISHER]

To fill the semantic gap between the predictive power of computational saliency models and human behavior, this paper proposes to predict where people look at using spatial-aware object-level cues. While object-level saliency has been recently suggested by psychophysics experiments and shown effective with a few computational models, the spatial relationship between the objects has not yet been explored in this context. We in this work for the first time explicitly model such spatial relationship, as well as leveraging semantic information of an image to enhance object-level saliency modeling. The core computational module is a graphlet-based (i.e., graphlets are moderate-sized connected subgraphs) deep architecture, which hierarchically learns a saliency map from raw image pixels to object-level graphlets (oGLs) and further to spatial-level graphlets (sGLs). Eye tracking data are also used to leverage human experience in saliency prediction. Experimental results demonstrate that the proposed oGLs and sGLs well capture object-level and spatial-level cues relating to saliency, and the resulting saliency model performs competitively compared with the state-of-the-art. [ABSTRACT FROM PUBLISHER]