Search

Skeleton-based action recognition, renowned for its computational efficiency and indifference to lighting variations, has become a focal point in the realm of motion analysis. However, most current methods typically only extract global skeleton features, overlooking the potential semantic relationships among various partial limb motions. For instance, the subtle differences between actions such as "brush teeth" and "brush hair" are mainly distinguished by specific elements. Although combining limb movements provides a more holistic representation of an action, relying solely on skeleton points proves inadequate for capturing these nuances. Therefore, integrating detailed linguistic descriptions into the learning process of skeleton features is essential. This motivates us to explore integrating fine-grained language descriptions into the learning process of skeleton features to capture more discriminative skeleton behavior representations. To this end, we introduce a new Linguistic-Driven Partial Semantic Relevance Learning framework (LPSR) in this work. While using state-of-the-art large language models to generate linguistic descriptions of local limb motions and further constrain the learning of local motions, we also aggregate global skeleton point representations and textual representations (which generated from an LLM) to obtain a more generalized cross-modal behavioral representation. On this basis, we propose a cyclic attentional interaction module to model the implicit correlations between partial limb motions. Numerous ablation experiments demonstrate the effectiveness of the method proposed in this paper, and our method also obtains state-of-the-art results. [ABSTRACT FROM AUTHOR]

RGB-Infrared person re-identification (Re-ID) is a challenging task which aims to match person images between visible and infrared modalities, playing a crucial role in criminal investigation and intelligent video surveillance. To address the weak feature extraction capability for fine-grained features in current cross-modal person Re-ID tasks, this paper proposes a person re-identification model based on fused attention and feature enhancement. First, automatic data augmentation techniques are employed to mitigate the differences in perspectives and scales among different cameras, and a cross-attention multi-scale Vision Transformer is proposed to generate more discriminative feature representations by processing multi-scale features. Then the channel attention and spatial attention mechanisms are introduced to learn information important for distinguishing features when fusing visible and infrared image features. Finally, a loss function is designed, which adopts the adaptive weight based hard triplet loss, to enhance the correlation between each sample and improve the capability of identifying different persons from visible and infrared images. Extensive experiments conducted on the SYSU-MM01 and RegDB datasets show that the proposed approach achieves mAP of 68.05% and 85.19%, respectively, outperforming many state-of-the-art approaches. Moreover, ablation experiments and comparative analysis validate the superiority and effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

With the rapid development of multimedia technology, considerable achievement has been achieved in image retrieval technology. On the one hand, users' demand for cross-modal retrieval is increasing rapidly. On the other hand, deep hashing algorithms as one of the most prominent high-dimensional reduction methods have received extensive attention. Under such background, the cross-modal retrieval method based on deep hashing came into being. Although the cross-modal hashing learning method has moved a long way in recent years, it still has space of promotion. First, existing cross-modal based hashing algorithms can only map the feature vectors into binary codes with a fixed length. The length of hash codes cannot be modified according to the practical situation. Second, the saturation level of the probability function used in existing methods is too high to concentrate relevant samples very well, which results in low retrieval efficiency. To solve the above problem, a novel variable-length hash code based on adaptive weight was proposed in this paper. The length of hash codes could be adjusted according to the importance of each bit. And a novel probability function based on Cauchy distribution was proposed to generate compact binary codes and make hashing retrieval more efficient. The experiment shows that the proposed cross-modal image retrieval algorithm based on deep hashing outperforms existing related algorithms on the accuracy of retrieval. [ABSTRACT FROM AUTHOR]

Cross-modal visible-infrared person re-identification (VI-ReID) aims to retrieve images of the same pedestrians captured by visible (VIS) cameras and infrared (IR) cameras and it is a challenging task in intelligent security systems. The differences in imaging principles between visible and infrared images lead to large cross-modal differences and intra-class differences, and such cross-modal image differences can be considered as special image style differences, while several intra-class differences can be considered as differences in the form of content expression between visible and infrared images. Some state-of-the-art methods improve the performance of the VI-ReID model by using additional feature enhancement or feature generation modules, however, these methods also introduce additional parameters and increase the training cost. In this paper, to mitigate the differences in image style and content between VIS and IR images, we design two objective functions based on content and style, which are style loss and content loss for the VI-ReID task, respectively. Our model can effectively mitigate the differences between modes by optimizing the objective function to map VIS and IR features into the same feature space without additional auxiliary modules. After extensive experiments, our model achieves competitive performance on two challenging datasets. Notably, under the visible2infrared setting on the RegDB dataset, our model achieves the state-of-the-art (SOTA) Rank-1/mAP/mINP = 96.13%/91.35%/83.67%. [ABSTRACT FROM AUTHOR]

The goal of RGB-D salient object detection is to aggregate the information of the two modalities of RGB and depth to accurately detect and segment salient objects. Existing RGB-D SOD models can extract the multilevel features of single modality well and can also integrate cross-modal features, but it can rarely handle both at the same time. To tap into and make the most of the correlations of intra- and inter-modality information, in this paper, we proposed an attention-guided cross-modal multi-feature aggregation network for RGB-D SOD. Our motivation was that both cross-modal feature fusion and multilevel feature fusion are crucial for RGB-D SOD task. The main innovation of this work lies in two points: One is the cross-modal pyramid feature interaction (CPFI) module that integrates multilevel features from both RGB and depth modalities in a bottom-up manner, and the other is cross-modal feature decoder (CMFD) that aggregates the fused features to generate the final saliency map. Extensive experiments on six benchmark datasets showed that the proposed attention-guided cross-modal multiple feature aggregation network (ACFPA-Net) achieved competitive performance over 15 state of the art (SOTA) RGB-D SOD methods, both qualitatively and quantitatively. [ABSTRACT FROM AUTHOR]

Cross-modal hashing which embeds data to binary codes is an efficient tool for retrieving heterogeneous but correlated multimedia data. In real applications, the sizes of queries are much larger than that of the training set and the queries may be dissimilar to training data, which lays bare the shortage of generalization of deterministic models, such as cross-encoder and autoencoder. In this paper, we design a variational cross-encoder (VCE), a generative model, to tackle this problem. At the bottleneck layer, the VCE outputs distributions parameterized by means and variances. As VCE can generate diversified data using noises, the proposed model can perform better in testing data. Ideally, each distribution is expected to describe a category of data and samples of this distribution are expected to generate data in the same category. Under this expectation, the means and variances can be used as real codes for input data. However, the generated data generally are not belonging to the same category as the input data. Hence, we add a penalty term on variance output of VCE and use means as real codes for further generating hashing codes. Experiments on three widely used datasets validate the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Learning about appearance embedding is of great importance for a variety of different computer-vision applications, which has prompted a surge in person re-identification (Re-ID) papers. The aim of these papers has been to identify an individual over a set of non-overlapping cameras. Despite recent advances in RGB–RGB Re-ID approaches with deep-learning architectures, the approach fails to consistently work well when there are low resolutions in dark conditions. The introduction of different sensors (i.e., RGB–D and infrared (IR)) enables the capture of appearances even in dark conditions. Recently, a lot of research has been dedicated to addressing the issue of finding appearance embedding in dark conditions using different advanced camera sensors. In this paper, we give a comprehensive overview of existing Re-ID approaches that utilize the additional information from different sensor-based methods to address the constraints faced by RGB camera-based person Re-ID systems. Although there are a number of survey papers that consider either the RGB–RGB or Visible-IR scenarios, there are none that consider both RGB–D and RGB–IR. In this paper, we present a detailed taxonomy of the existing approaches along with the existing RGB–D and RGB–IR person Re-ID datasets. Then, we summarize the performance of state-of-the-art methods on several representative RGB–D and RGB–IR datasets. Finally, future directions and current issues are considered for improving the different sensor-based person Re-ID systems. [ABSTRACT FROM AUTHOR]

Dogs can recognize conspecifics in cross-modal audio–video presentations. In this paper, we aimed at exploring if such capability extends to the recognition of cats, and whether it is influenced by exposure to these animals. To reach our aim, we enrolled 64 pet dogs. Half of the dogs were currently living with cats, while the rest had never been living with cats, nor were at the time of the experiment. All dogs underwent a cross-modal violation of expectancy experiment, where they were presented with either a cat or a dog vocalization, followed by a video of either species on a blank background. The result revealed that dogs did not exhibit a surprise reaction towards the incoherent stimuli of a cat vocalization and a dog video or vice-versa, implying that they had not recognized the stimuli portraying cats. The pattern of results did not differ between dogs living or not with cats, implying that exposure to a limited number of cats, however, prolonged, is not sufficient to grant dogs with the ability to recognize them on audio–video presentations. We propose that the lack of recognition could be due to the small number of individual cats the dogs are regularly exposed to, or to the possible lack of early exposure to cats during the socialization phase. [ABSTRACT FROM AUTHOR]

Inspired by visual-tactile cross-modal bidirectional mapping of the human brain, this paper introduces a novel approach to bidirectional mapping between visual and tactile data, an area not fully explored in the predominantly unidirectional existing studies. First, we adopt separate Variational AutoEncoder (VAE) models for visual and tactile data. Furthermore, we introduce a conditional flow model built on the VAE latent feature space, enabling cross-modal bidirectional mapping between visual and tactile data using one model. The experimental results show that our method achieves excellent performance in terms of the similarity between the generated data and the original data (Structural Similarity Index (SSIM) of visual data: 0.58, SSIM of tactile data: 0.80), the classification accuracy on generated data (visual data: 91.60%, tactile data: 88.05%), and the zero-shot classification accuracy between generated data and language (visual data: 44.49%, tactile data: 45.03%). To the best of our knowledge, the method proposed in this paper is the first one to utilize a single model to achieve bidirectional mapping between visual and tactile data. Our model and code will be made public after the acceptance of the paper. [ABSTRACT FROM AUTHOR]

We consider the problem of referring segmentation in images and videos with natural language. Given an input image (or video) and a referring expression, the goal is to segment the entity referred by the expression in the image or video. In this paper, we propose a cross-modal self-attention (CMSA) module to utilize fine details of individual words and the input image or video, which effectively captures the long-range dependencies between linguistic and visual features. Our model can adaptively focus on informative words in the referring expression and important regions in the visual input. We further propose a gated multi-level fusion (GMLF) module to selectively integrate self-attentive cross-modal features corresponding to different levels of visual features. This module controls the feature fusion of information flow of features at different levels with high-level and low-level semantic information related to different attentive words. Besides, we introduce cross-frame self-attention (CFSA) module to effectively integrate temporal information in consecutive frames which extends our method in the case of referring segmentation in videos. Experiments on benchmark datasets of four referring image datasets and two actor and action video segmentation datasets consistently demonstrate that our proposed approach outperforms existing state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Salient object detection (SOD) aims at highlighting important foreground objects automatically from the background. Most existing SOD methods only employ visible images (RGB images) for salient detection, which limits the performance of real-life applications when encountering challenging scenarios such as low illumination, haze, and smog. In this paper, we take advantage of the RGB and thermal images and propose an Enhancement and Aggregation–Feedback Network (EAF-Net) for SOD. Specifically, to achieve effective complementation between modalities and prevent the interference from noises, we first treat RGB and thermal images equally in the Feature Enhancement Block (FEB), and further, the Global Context Module expands receptive field to obtain the global features and the Top-Feature Enhancement Module suppresses the redundant information that may destroy the original features from the top layer. Subsequently, we embed several Cross Feature Aggregation Modules (CFAMs) into the Aggregation-and-Feedback Decoder to fuse different level features and compensation features for further obtaining comprehensive feature expression. Moreover, a feedback mechanism is adopted to propagate these fused features back into previous layers for refinement and generate saliency maps to decode features in a progressive way. Comprehensive experiments on RGB-T datasets demonstrate that EAF-Net achieves outstanding performance against the state-of-the-art models. [ABSTRACT FROM AUTHOR]

Recent progress in generative models has led to the drastic growth of research in image generation. Existing approaches show visually compelling results by learning multi-modal distributions, but they still lack realism, especially in certain scenarios like medical image synthesis. In this paper, we propose a novel Brain Generative Adversarial Network (BrainGAN) that explores GANs with multi-constraint and transferable property for cross-modal brain image synthesis. We formulate BrainGAN by introducing a unified framework with new constraints that can enhance modal matching, texture details and anatomical structure, simultaneously. We show how BrainGAN can learn meaningful tissue representations with rich variability of brain images. In addition to generating 3D volumes that are visually indistinguishable from real ones, we model adversarial discriminators and segmentors jointly, along with the proposed cost functions, which forces our networks to synthesize brain MRIs with realistic textures conditioned on anatomical structures. BrainGAN is evaluated on three public datasets, where it consistently outperforms the other state-of-the-art approaches by a large margin, advancing cross-modal synthesis of brain images both visually and practically. [ABSTRACT FROM AUTHOR]

With the ever-increasing urbanization process, modeling people's spatiotemporal activities from their online traces has become a crucial task. State-of-the-art methods for this task rely on cross-modal embedding, which maps items from different modalities (e.g., location, time, text) into the same latent space. Despite their inspiring results, existing cross-modal embedding methods merely capture co-occurrences between items without modeling their high-order interactions. In this paper, we first construct two graphs from raw data records to represent the user interaction graph layer and activity graph layer and propose a hierarchical cross-modal embedding method that takes the high-order relationships into consideration. The key notion behind our method is a novel hierarchical embedding framework with meta-graphs connecting different layers. We introduce both inter-record and intra-record meta-graph structures, which enable learning distributed representations that preserve high-order proximities across graphs from different layers. Our empirical experiments on three real-world datasets demonstrate that our method not only outperforms state-of-the-art methods for spatiotemporal activity prediction, but also captures cross-modal proximity at a finer granularity. [ABSTRACT FROM AUTHOR]

• We propose Cross-Modal Transformer (CMFormer), a Transformer-based cross-modal semantic segmentation model, which achieves better cross-modal information interaction by capturing long-range contextual dependencies. The CMFormer includes the MS-CAC module and the GFA module, and achieves the state-of-the-art results in RGB-D semantic segmentation task. • We design the MS-CAC module and the GFA module by exploiting the cross-modal properties of Transformer. The MS-CAC module preserves multi-scale information while realizing channel feature correction, while the GFA module realizes spatial feature fusion, and fully considers global and local features, which make the CMFormer achieve better cross-modal information interaction by capturing long-range contextual dependencies. • The effectiveness of the CMFormer is evaluated with extensive experiments on the SOP dataset and NYU Depth v2 dataset. The results show that the CMFormer achieves the state-of-the-art results on both datasets. In the experiment of SOP dataset, CMFormer can achieve 96.74% MPA and 92.98% mIoU. At the same time, the real-time performance of CMFormer can reach 43 FPS(Frames Per Second), which has good real-time performance while meeting the requirements of high precision. Scene understanding in a production workshop is an important technology to improve its intelligence level, semantic segmentation of production workshop objects is an effective method for realizing scene understanding. Since the varieties of information of production workshop, making full use of the complementary information of RGB image and depth image can effectively improve the semantic segmentation accuracy of production workshop objects. Aiming at solving the multi-scale and real-time problems of segmenting the production workshop objects, this paper proposes Cross-Modal Transformer (CMFormer), a Transformer-based cross-modal semantic segmentation model. Its key feature correction and feature fusion parts are composed of the Multi-Scale Channel Attention Correction(MS-CAC) module and the Global Feature Aggregation(GFA) module. By improving Multi-Head Self-Attention(MHSA) in Transformer, we design Cross-Modal Multi-Head Self-Attention(CM-MHSA) to build long-range interaction between RGB image and depth image, and further design the MS-CAC module and the GFA module on the basis of the CM-MHSA module, to achieve cross-modal information interaction in the channel and spatial dimensions. Among them, the MS-CAC module enriches the multi-scale features of each channel and achieve more accurate channel attention correction between the two modals; the GFA module interacts with RGB feature and depth feature in the spatial dimension and fuses global and local features at the same time. In the experiments on the NYU Depth v2 dataset, the CMFormer reached 68.00% MPA(Mean Pixel Accuracy) and 55.75% mIoU(Mean Intersection over Union), achieves the state-of-the-art results. While in the experiments on the Scene Objects for Production workshop dataset(SOP), t he CMFormer achieves 96.74% MPA, 92.98% mIoU and 43 FPS(Frames Per Second), which has high precision and good real-time performance. Code is available at: https://github.com/FutureIAI/CMFormer [ABSTRACT FROM AUTHOR]

• Existing Visual Semantic Embedding (VSE) networks are trained by a hard negatives loss function that learns an objective margin between the similarity of relevant and irrelevant image–description embedding pairs and ignores the semantic differences between the irrelevant pairs. • We propose a novel Semantically Enhanced Hard negatives Loss function (LSEH) for Cross-modal Information Retrieval that considers the semantic differences between irrelevant training pairs. • The proposed LSEH function dynamically adjusts the learning objectives of VSE networks to make their learning flexible and efficient. • Experiments with various benchmark datasets and VSE networks revealed that the proposed LSEH function reduces their training epochs by approximately 50% and also improves their retrieval performance. Visual Semantic Embedding (VSE) networks aim to extract the semantics of images and their descriptions and embed them into the same latent space for cross-modal information retrieval. Most existing VSE networks are trained by adopting a hard negatives loss function which learns an objective margin between the similarity of relevant and irrelevant image–description embedding pairs. However, the objective margin in the hard negatives loss function is set as a fixed hyperparameter that ignores the semantic differences of the irrelevant image–description pairs. To address the challenge of measuring the optimal similarities between image–description pairs before obtaining the trained VSE networks, this paper presents a novel approach that comprises two main parts: (1) finds the underlying semantics of image descriptions; and (2) proposes a novel semantically-enhanced hard negatives loss function, where the learning objective is dynamically determined based on the optimal similarity scores between irrelevant image–description pairs. Extensive experiments were carried out by integrating the proposed methods into five state-of-the-art VSE networks that were applied to three benchmark datasets for cross-modal information retrieval tasks. The results revealed that the proposed methods achieved the best performance and can also be adopted by existing and future VSE networks. [ABSTRACT FROM AUTHOR]

RGB-D saliency detection aims to identify the most attractive objects in a pair of color and depth images. However, most existing models adopt classic U-Net framework which progressively decodes two-stream features. In this paper, we decode the cross-modal and multi-level features in a unified unit, named Attention Gated Recurrent Unit (AGRU). It can reduce the influence of low-quality depth image, and retain more semantic features in the progressive fusion process. Specifically, the features of different modalities and different levels are organized as the sequential input, recurrently fed into AGRU which consists of reset gate, update gate and memory unit to be selectively fused and adaptively memorized based on attention mechanism. Further, two-stage AGRU serves as the decoder of RGB-D salient object detection network, named AGRFNet. Due to the recurrent nature, it achieves the best performance with the little parameters. In order to further improve the performance, three auxiliary modules are designed to better fuse semantic information, refine the features of the shallow layer and enhance the local detail. Extensive experiments on seven widely used benchmark datasets demonstrate that AGRFNet performs favorably against 18 state-of-the-art RGB-D SOD approaches. • A cross-modal and multi-level attention gated recurrent fusion network is proposed for salient object detection in RGB-D image. It uses two-stage gated recurrent unit to fuse the cross-modal and multi-level features in the decoding process. • Attention GRU (AGRU) is proposed to enhance the cross-modal fusion or multi-level fusion in a unified structure based on attention mechanism. It can selectively combine color and depth image, and adaptively remember optimal fusion result. • To further improve the performance of our network, three modules (CFM, LRM, DEM) are proposed. They are designed to fuse the high-layer semantic information, refine the low-layer features, and enhance the local details. [ABSTRACT FROM AUTHOR]