Search

The olfactory system enables humans to smell different odors, which are closely related to emotions. The high temporal resolution and non-invasiveness of Electroencephalogram (EEG) make it suitable to objectively study human preferences for odors. Effectively learning the temporal dynamics and spatial information from EEG is crucial for detecting odor-induced emotional valence. In this paper, we propose a deep learning architecture called Temporal Attention with Spatial Autoencoder Network (TASA) for predicting odor-induced emotions using EEG. TASA consists of a filter-bank layer, a spatial encoder, a time segmentation layer, a Long Short-Term Memory (LSTM) module, a multi-head self-attention (MSA) layer, and a fully connected layer. We improve upon the previous work by utilizing a two-phase learning framework, using the autoencoder module to learn the spatial information among electrodes by reconstructing the given input with a latent representation in the spatial dimension, which aims to minimize information loss compared to spatial filtering with CNN. The second improvement is inspired by the continuous nature of the olfactory process; we propose to use LSTM-MSA in TASA to capture its temporal dynamics by learning the intercorrelation among the time segments of the EEG. TASA is evaluated on an existing olfactory EEG dataset and compared with several existing deep learning architectures to demonstrate its effectiveness in predicting olfactory-triggered emotional responses. Interpretability analyses with DeepLIFT also suggest that TASA learns spatial-spectral features that are relevant to olfactory-induced emotion recognition.

As an essential cognitive function, attention has been widely studied and various indices based on EEG have been proposed for its convenience and easy availability for real-time attention monitoring. Although existing indices based on spectral power of empirical frequency bands are able to describe the attentional state in some way, the reliability still needs to be improved. This paper proposed a subject-specific attention index based on the weighted spectral power. Unlike traditional indices, the ranges of frequency bands are not empirical but obtained from subject-specific change patterns of spectral power of electroencephalograph (EEG) to overcome the great inter-subject variance. In addition, the contribution of each frequency component in the frequency band is considered different. Specifically, the ratio of power spectral density (PSD) function in attentional and inattentional state is utilized to calculate the weight to enhance the effectiveness of the proposed index. The proposed subject-specific attention index based on the weighted spectral power is evaluated on two open datasets including EEG data of a total of 44 subjects. The results of the proposed index are compared with 3 traditional attention indices using various statistical analysis methods including significance tests and distribution variance measurements. According to the experimental results, the proposed index can describe the attentional state more accurately. The proposed index respectively achieves accuracies of 86.21% and 70.00% at the 1% significance level in both the t-test and Wilcoxon rank-sum test for two datasets, which obtains improvements of 41.38% and 20.00% compared to the best result of the traditional indices. These results indicate that the proposed index provides an efficient way to measure attentional state.

Domain adaptation aims to alleviate the distribution discrepancy between source and target domains. Most conventional methods focus on one target domain setting adapted from one or multiple source domains while neglecting the multi-target domain setting. We argue that different target domains also have complementary information, which is very important for performance improvement. In this paper, we propose an Attention-guided Multiple source-and-target Domain Adaptation (AMDA) method to capture the context dependency information on transferable regions among multiple source and target domains. The innovation points of this paper are as follows: (1) We use numerous adversarial strategies to harvest sufficient information from multiple source and target domains, which extends the generalization and robustness of the feature pools. (2) We propose an intra-domain and inter-domain attention module to explore transferable context information. The proposed attention module can learn domain-invariant representations and reduce the negative transfer by focusing on transferable knowledge. Extensive experiments validate the effectiveness of our method with achieving state-of-the-art performance on several unsupervised domain adaptation datasets. [ABSTRACT FROM AUTHOR]

As a kind of non-invasive, low-cost, and readily available brain examination, EEG has attached significance to the means of clinical diagnosis of epilepsy. However, the reading of long-term EEG records has brought a heavy burden to neurologists and experts. Therefore, automatic EEG classification for epileptic patients plays an essential role in epilepsy diagnosis and treatment. This paper proposes an Attention Mechanism-based Wavelet Convolution Neural Network for epilepsy EEG classification. Attention Mechanism-based Wavelet Convolution Neural Network firstly uses multi-scale wavelet analysis to decompose the input EEGs to obtain their components in different frequency bands. Then, these decomposed multi-scale EEGs are input into the Convolution Neural Network with an attention mechanism for further feature extraction and classification. The proposed algorithm achieves 98.89% triple classification accuracy on the Bonn EEG database and 99.70% binary classification accuracy on the Bern-Barcelona EEG database. Our experiments prove that the proposed algorithm achieves a state-of-the-art classification effect on epilepsy EEG. [ABSTRACT FROM AUTHOR]

Recently, substantial research effort has focused on how to apply CNNs or RNNs to better capture temporal patterns in videos, so as to improve the accuracy of video classification. In this paper, we investigate the potential of a purely attention based local feature integration. Accounting for the characteristics of such features in video classification, we first propose Basic Attention Clusters (BAC), which concatenates the output of multiple attention units applied in parallel, and introduce a shifting operation to capture more diverse signals. Experiments show that BAC can achieve excellent results on multiple datasets. However, BAC treats all feature channels as an indivisible whole, which is suboptimal for achieving a finer-grained local feature integration over the channel dimension. Additionally, it treats the entire local feature sequence as an unordered set, thus ignoring the sequential relationships. To improve over BAC, we further propose the channel pyramid attention schema by splitting features into sub-features at multiple scales for coarse-to-fine sub-feature interaction modeling, and propose the temporal pyramid attention schema by dividing the feature sequences into ordered sub-sequences of multiple lengths to account for the sequential order. Our final model pyramid×pyramid attention clusters (PPAC) combines both channel pyramid attention and temporal pyramid attention to focus on the most important sub-features, while also preserving the temporal information of the video. We demonstrate the effectiveness of PPAC on seven real-world video classification datasets. Our model achieves competitive results across all of these, showing that our proposed framework can consistently outperform the existing local feature integration methods across a range of different scenarios. [ABSTRACT FROM AUTHOR]

Scene classification of remote sensing images has drawn great attention because of its wide applications. In this paper, with the guidance of the human visual system (HVS), we explore the attention mechanism and propose a novel end-to-end attention recurrent convolutional network (ARCNet) for scene classification. It can learn to focus selectively on some key regions or locations and just process them at high-level features, thereby discarding the noncritical information and promoting the classification performance. The contributions of this paper are threefold. First, we design a novel recurrent attention structure to squeeze high-level semantic and spatial features into several simplex vectors for the reduction of learning parameters. Second, an end-to-end network named ARCNet is proposed to adaptively select a series of attention regions and then to generate powerful predictions by learning to process them sequentially. Third, we construct a new data set named OPTIMAL-31, which contains more categories than popular data sets and gives researchers an extra platform to validate their algorithms. The experimental results demonstrate that our model makes great promotion in comparison with the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Decoding the motion intention of the human upper limb from electroencephalography (EEG) signals has important practical values. However, existing decoding models are built under the attended state while subjects perform motion tasks. In practice, people are often distracted by other tasks or environmental factors, which may impair decoding performance. To address this problem, in this paper, we propose a hierarchical decoding model of human upper limb motion intention from EEG signals based on attention state estimation. The proposed decoding model includes two components. First, the attention state detection (ASD) component estimates the attention state during the upper limb movement. Next, the motion intention recognition (MIR) component decodes the motion intention by using the decoding models built under the attended and distracted states. The experimental results show that the proposed hierarchical decoding model performs well under the attended and distracted states. This work can advance the application of human movement intention decoding and provides new insights into the study of brain-machine interfaces. [ABSTRACT FROM AUTHOR]

Unsupervised domain adaptation crowd counting (UDACC) has been studied with practical research utility by getting rid of the labeling burden on large-scale dense crowds in the target domain. Current methods generalize well within the specific domain gap by directly aligning domain distributions or translating synthetic data to realistic images. However, it is difficult to define domain gaps among complex real-world datasets, in which the images vary greatly in style, density level and/or content. To tackle this problem, in this paper, we propose a Cross-Domain Attention Network (CDANet), which can effectively generalize the model to the unlabeled domain on both unsupervised synthetic-to-realistic and realistic-to-realistic crowd counting. Specifically, we propose a Cross-Domain Attention Module (CDAM) to learn domain-related information between the source and target domain, which extracts relations in cross-domain attentive information, thus enhancing crowd-informative features. Moreover, to make our CDAM invariant to domain shifts, we introduce a consistency penalty to ensure that the attention maps are consistent before and after the domain shifting. Thus our CDANet can pay attention to the shared counting information across domains, while remaining its invariant ability during domain adaptation. Extensive experiments on several common benchmarks for UDACC demonstrate that our CDANet gets competitive results on both unsupervised synthetic-to-realistic and realistic-to-realistic UDACC tasks. [ABSTRACT FROM AUTHOR]

Attention mechanisms have significantly boosted the performance of video classification neural networks thanks to the utilization of perspective contexts. However, the current research on video attention generally focuses on adopting a specific aspect of contexts (e.g., channel, spatial/temporal, or global context) to refine the features and neglects their underlying correlation when computing attentions. This leads to incomplete context utilization and hence bears the weakness of limited performance improvement. To tackle the problem, this paper proposes an efficient attention-in-attention (AIA) method for element-wise feature refinement, which investigates the feasibility of inserting the channel context into the spatio-temporal attention learning module, referred to as CinST, and also its reverse variant, referred to as STinC. Specifically, we instantiate the video feature contexts as dynamics aggregated along a specific axis with global average and max pooling operations. The workflow of an AIA module is that the first attention block uses one kind of context information to guide the gating weights calculation of the second attention that targets at the other context. Moreover, all the computational operations in attention units act on the pooled dimension, which results in quite few computational cost increase (< 0.02%). To verify our method, we densely integrate it into two classical video network backbones and conduct extensive experiments on several standard video classification benchmarks. The source code of our AIA is available at https://github.com/haoyanbin918/Attention-in-Attention. [ABSTRACT FROM AUTHOR]

Human beings can concentrate on the most semantically relevant visual information when performing action recognition, so as to make reasonable and interpretable predictions. However, most existing approaches, which are applied to address visual tasks, neglect to explicitly imitate such ability for improving the performance and reliability of models. In this paper, we propose an interpretable action recognition framework that can not only improve the performance but also enhance the visual interpretability of 3D CNNs. Specifically, we design a semantic-aware attention module to learn correlative spatial-temporal attention for different action categories. To further leverage the rich semantics of features extracted from different layers, we design a hierarchical semantic fusion module with the help of the learned attention. The proposed two modules can enhance and complement each other, meanwhile, the semantic-aware attention module enjoys the plug-and-play merit. We evaluate our method on different benchmarks with comprehensive ablation studies and visualization analysis. Experimental results demonstrate the effectiveness of our method, showing favorable accuracy against state-of-the-arts while enhancing the semantic interpretability (Code will be available at this link https://github.com/PHDJieFu). [ABSTRACT FROM AUTHOR]

Remarkable achievements have been obtained by deep neural networks in the last several years. However, the breakthrough in neural networks accuracy is always accompanied by explosive growth of computation and parameters, which leads to a severe limitation of model deployment. In this paper, we propose a novel knowledge distillation technique named self-distillation to address this problem. Self-distillation attaches several attention modules and shallow classifiers at different depths of neural networks and distills knowledge from the deepest classifier to the shallower classifiers. Different from the conventional knowledge distillation methods where the knowledge of the teacher model is transferred to another student model, self-distillation can be considered as knowledge transfer in the same model - from the deeper layers to the shallow layers. Moreover, the additional classifiers in self-distillation allow the neural network to work in a dynamic manner, which leads to a much higher acceleration. Experiments demonstrate that self-distillation has consistent and significant effectiveness on various neural networks and datasets. On average, 3.49 and 2.32 percent accuracy boost are observed on CIFAR100 and ImageNet. Besides, experiments show that self-distillation can be combined with other model compression methods, including knowledge distillation, pruning and lightweight model design. [ABSTRACT FROM AUTHOR]

Stereopsis is the ability of human beings to get the 3D perception on real scenarios. The conventional stereopsis measurement is based on subjective judgment for stereograms, leading to be easily affected by personal consciousness. To alleviate the issue, in this paper, the EEG signals evoked by dynamic random dot stereograms (DRDS) are collected for stereogram recognition, which can help the ophthalmologists diagnose strabismus patients even without real-time communication. To classify the collected Electroencephalography (EEG) signals, a novel multi-scale temporal self-attention and dynamical graph convolution hybrid network (MTS-DGCHN) is proposed, including multi-scale temporal self-attention module, dynamical graph convolution module and classification module. Firstly, the multi-scale temporal self-attention module is employed to learn time continuity information, where the temporal self-attention block is designed to highlight the global importance of each time segments in one EEG trial, and the multi-scale convolution block is developed to further extract advanced temporal features in multiple receptive fields. Meanwhile, the dynamical graph convolution module is utilized to capture spatial functional relationships between different EEG electrodes, in which the adjacency matrix of each GCN layer is adaptively tuned to explore the optimal intrinsic relationship. Finally, the temporal and spatial features are fed into the classification module to obtain prediction results. Extensive experiments are conducted on collected datasets i.e., SRDA and SRDB, and the results demonstrate the proposed MTS-DGCHN achieves outstanding classification performance compared with the other methods. The datasets are available at https://github.com/YANGeeg/TJU-SRD-datasets and the code is at https://github.com/YANGeeg/MTS-DGCHN.

By involving only subtle motions of body parts, video-based microaction recognition is a very important but challenging problem. Most existing action recognition methods are developed for general actions, and the current state-of-the-art methods usually largely rely on high-layer features learned from convolutional neural networks (CNNs). High-layer CNN features usually contain more semantic information but less detailed information. However, detailed information can be important for microactions due to the motion subtleness of such actions. In this paper, we propose to more effectively learn midlayer CNN features for enhancing microaction recognition. More specifically, we develop a new dual-branch network for microaction recognition: one branch uses the high-layer CNN features for classification, and the second branch further explores the midlayer CNN features for classification. In the second branch, we introduce a novel subtle motion detector consisting of three modules: 1) a discriminative spatial-temporal feature learning module, which further learns the subtle motion features corresponding to the discriminative spatial-temporal regions, 2) a parallel multiplier attention module, which further refines the features learned in channels and spatial-temporal domains, and 3) an activation fusion module, which fuses the max and average activations from midlayer CNN features for classification. In the experiments, we build a new microaction video dataset, where the micromotions of interest are mixed with other larger general motions such as walking. Comprehensive experimental results verify that the proposed method yields new state-of-the-art performance in two microaction video datasets, while its performance on two general-action video datasets is also very promising. [ABSTRACT FROM AUTHOR]

Most existing methods of constructing dynamic functional connectivity (dFC) network obtain the connectivity strength via the sliding window correlation (SWC) method, which estimates the connectivity strength at each time segment, rather than at each time point, and thus is difficult to produce accurate dFC network due to the influence of the window type and window width. Furthermore, the deep learning methods may not capture the discriminative spatio-temporal information that is closely related to disease, thus impacting the performance of mild cognitive impairment (MCI) identification. In this paper, a novel spatio-temporal attention-based bidirectional gated recurrent unit (STA-BiGRU) network is proposed to extract inherent spatio-temporal information from a dynamic adaptive functional connectivity (dAFC) network for MCI diagnosis. Specifically, we adopt a group lasso-based Kalman filter algorithm to obtain the dAFC network with more accurate connectivity strength at each time step. Then a spatial attention module with self-attention and a temporal attention module with multiple temporal attention vectors are incorporated into the BiGRU network to extract more discriminative disease-related spatio-temporal information. Finally, the spatio-temporal regularizations are employed to better guide the attention learning of STA-BiGRU network to enhance the robustness of the deep network. Experimental results show that the proposed framework achieves mean accuracies of 90.2%, 90.0%, and 81.5%, respectively, for three MCI classification tasks. This study provides a more effective deep spatio-temporal attention-based recurrent network and obtains good performance and interpretability of deep learning for psychiatry diagnosis research. [ABSTRACT FROM AUTHOR]

Automatically recognising surgical gestures from surgical data is an important building block of automated activity recognition and analytics, technical skill assessment, intra-operative assistance and eventually robotic automation. The complexity of articulated instrument trajectories and the inherent variability due to surgical style and patient anatomy make analysis and fine-grained segmentation of surgical motion patterns from robot kinematics alone very difficult. Surgical video provides crucial information from the surgical site with context for the kinematic data and the interaction between the instruments and tissue. Yet sensor fusion between the robot data and surgical video stream is non-trivial because the data have different frequency, dimensions and discriminative capability. In this paper, we integrate multimodal attention mechanisms in a two-stream temporal convolutional network to compute relevance scores and weight kinematic and visual feature representations dynamically in time, aiming to aid multimodal network training and achieve effective sensor fusion. We report the results of our system on the JIGSAWS benchmark dataset and on a new in vivo dataset of suturing segments from robotic prostatectomy procedures. Our results are promising and obtain multimodal prediction sequences with higher accuracy and better temporal structure than corresponding unimodal solutions. Visualization of attention scores also gives physically interpretable insights on network understanding of strengths and weaknesses of each sensor. [ABSTRACT FROM AUTHOR]

In leveraging manifold learning in domain adaptation (DA), graph embedding-based DA methods have shown their effectiveness in preserving data manifold through the Laplace graph. However, current graph embedding DA methods suffer from two issues: 1). they are only concerned with preservation of the underlying data structures in the embedding and ignore sub-domain adaptation, which requires taking into account intra-class similarity and inter-class dissimilarity, thereby leading to negative transfer; 2). manifold learning is proposed across different feature/label spaces separately, thereby hindering unified comprehensive manifold learning. In this paper, starting from our previous DGA-DA, we propose a novel DA method, namely ${A}$ ttention ${R}$ egularized Laplace ${G}$ raph-based ${D}$ omain ${A}$ daptation (ARG-DA), to remedy the aforementioned issues. Specifically, by weighting the importance across different sub-domain adaptation tasks, we propose the ${A}$ ttention ${R}$ egularized Laplace Graph for class aware DA, thereby generating the attention regularized DA. Furthermore, using a specifically designed FEEL strategy, our approach dynamically unifies alignment of the manifold structures across different feature/label spaces, thus leading to comprehensive manifold learning. Comprehensive experiments are carried out to verify the effectiveness of the proposed DA method, which consistently outperforms the state of the art DA methods on 7 standard DA benchmarks, i.e., 37 cross-domain image classification tasks including object, face, and digit images. An in-depth analysis of the proposed DA method is also discussed, including sensitivity, convergence, and robustness. [ABSTRACT FROM AUTHOR]

Due to the wavelength-dependent light attenuation, refraction and scattering, underwater images usually suffer from color distortion and blurred details. However, due to the limited number of paired underwater images with undistorted images as reference, training deep enhancement models for diverse degradation types is quite difficult. To boost the performance of data-driven approaches, it is essential to establish more effective learning mechanisms that mine richer supervised information from limited training sample resources. In this paper, we propose a novel underwater image enhancement network, called SGUIE-Net, in which we introduce semantic information as high-level guidance via region-wise enhancement feature learning. Accordingly, we propose semantic region-wise enhancement module to better learn local enhancement features for semantic regions with multi-scale perception. After using them as complementary features and feeding them to the main branch, which extracts the global enhancement features on the original image scale, the fused features bring semantically consistent and visually superior enhancements. Extensive experiments on the publicly available datasets and our proposed dataset demonstrate the impressive performance of SGUIE-Net. The code and proposed dataset are available at https://trentqq.github.io/SGUIE-Net.html . [ABSTRACT FROM AUTHOR]

In this paper, we study the cross-view geo-localization problem to match images from different viewpoints. The key motivation underpinning this task is to learn a discriminative viewpoint-invariant visual representation. Inspired by the human visual system for mining local patterns, we propose a new framework called RK-Net to jointly learn the discriminative Representation and detect salient Keypoints with a single Network. Specifically, we introduce a Unit Subtraction Attention Module (USAM) that can automatically discover representative keypoints from feature maps and draw attention to the salient regions. USAM contains very few learning parameters but yields significant performance improvement and can be easily plugged into different networks. We demonstrate through extensive experiments that (1) by incorporating USAM, RK-Net facilitates end-to-end joint learning without the prerequisite of extra annotations. Representation learning and keypoint detection are two highly-related tasks. Representation learning aids keypoint detection. Keypoint detection, in turn, enriches the model capability against large appearance changes caused by viewpoint variants. (2) USAM is easy to implement and can be integrated with existing methods, further improving the state-of-the-art performance. We achieve competitive geo-localization accuracy on three challenging datasets, i. e., University-1652, CVUSA and CVACT. Our code is available at https://github.com/AggMan96/RK-Net. [ABSTRACT FROM AUTHOR]

Recently, siamese-based trackers have achieved significant successes. However, those trackers are restricted by the difficulty of learning consistent feature representation with the object. To address the above challenge, this paper proposes a novel siamese implicit region proposal network with compound attention for visual tracking. First, an implicit region proposal (IRP) module is designed by combining a novel pixel-wise correlation method. This module can aggregate feature information of different regions that are similar to the pre-defined anchor boxes in Region Proposal Network. To this end, the adaptive feature receptive fields then can be obtained by linear fusion of features from different regions. Second, a compound attention module including a channel and non-local attention is raised to assist the IRP module to perform a better perception of the scale and shape of the object. The channel attention is applied for mining the discriminative information of the object to handle the background clutters of the template, while non-local attention is trained to aggregate the contextual information to learn the semantic range of the object. Finally, experimental results demonstrate that the proposed tracker achieves state-of-the-art performance on six challenging benchmark tests, including VOT-2018, VOT-2019, OTB-100, GOT-10k, LaSOT, and TrackingNet. Further, our obtained results demonstrate that the proposed approach can be run at an average speed of 72 FPS in real time. [ABSTRACT FROM AUTHOR]

Real-world scenes always exhibit objects with clutter backgrounds, posing great challenges for deep salient object detection models. In this paper, we propose salient object detection by engaging two saliency cues, i.e., the part-whole hierarchies and contrast cues, resulting in a PWHCNet. Specifically, two branches, which consists of a Dynamic Grouping Capsules (DGC) branch and a DenseHRNet branch, are put in place to learn the part-whole hierarchies and contrast cues, respectively. Moreover, to help highlight the whole salient object in complex scenes, a Background Suppression (BS) module is proposed to guide the shallow features of DenseHRNet with the aid of the part-whole relational cues captured by DGC. Subsequently, these two saliency cues are integrated via a Self-Channel and Mutual-Spatial (SCMS) attention mechanism. Experimental results on five benchmarks demonstrate that the proposed PWHCNet achieves state-of-the-art performance while obtaining the whole salient objects with fine details. [ABSTRACT FROM AUTHOR]

Human mobility prediction is of great importance for a wide spectrum of location-based applications. However, predicting mobility is not trivial because of four challenges: 1) the complex sequential transition regularities exhibited with time-dependent and high-order nature; 2) the multi-level periodicity of human mobility; 3) the heterogeneity and sparsity of the collected trajectory data; and 4) the complicated semantic motivation behind the mobility. In this paper, we propose DeepMove, an attentional recurrent network for mobility prediction from lengthy and sparse trajectories. In DeepMove, we first design a multi-modal embedding recurrent neural network to capture the complicated sequential transitions by jointly embedding the multiple factors that govern human mobility. Then, we propose a historical attention model with two mechanisms to capture the multi-level periodicity in a principle way, which effectively utilizes the periodicity nature to augment the recurrent neural network for mobility prediction. Furthermore, we design a context adaptor to capture the semantic effects of Point-Of-Interest (POI)-based activity and temporal factor (e.g., dwell time). Finally, we use the multi-task framework to encourage the model to learn comprehensive motivations with mobility by introducing the task of the next activity type prediction and the next check-in time prediction. We perform experiments on four representative real-life mobility datasets, and extensive evaluation results demonstrate that our model outperforms the state-of-the-art models by more than 10 percent. Moreover, compared with the state-of-the-art neural network models, DeepMove provides intuitive explanations into the prediction and sheds light on interpretable mobility prediction. [ABSTRACT FROM AUTHOR]

Personalized location prediction is key to many mobile applications and services. In this paper, motivated by both statistical and visualized preliminary analysis on three real datasets, we observe a strong spatiotemporal correlation for user trajectories among the visited area-of-interests (AoIs) and different time periods on both weekly and daily basis, which directly motivates our time-aware location prediction model design called “ $t$ t -LocPred”. It models the spatial correlations among AoIs by coarse-grained convolutional processing of the user trajectories in AoIs of different time periods (“ConvAoI”); and predicts his/her fine-grained next visited PoI using a novel memory-augmented attentive LSTM model (“mem-attLSTM”) to capture long-term behavior patterns. Experimental results show that $t$ t -LocPred outperforms 8 baselines. We also show the impact of hyperparameters and the benefits ConvAoI can bring to these baselines. [ABSTRACT FROM AUTHOR]

Adversarial attacks on deep neural networks (DNNs) have been found for several years. However, the existing adversarial attacks have high success rates only when the information of the victim DNN is well-known or could be estimated by the structure similarity or massive queries. In this paper, we propose to Attack on Attention (AoA), a semantic property commonly shared by DNNs. AoA enjoys a significant increase in transferability when the traditional cross entropy loss is replaced with the attention loss. Since AoA alters the loss function only, it could be easily combined with other transferability-enhancement techniques and then achieve SOTA performance. We apply AoA to generate 50000 adversarial samples from ImageNet validation set to defeat many neural networks, and thus name the dataset as DAmageNet. 13 well-trained DNNs are tested on DAmageNet, and all of them have an error rate over 85 percent. Even with defenses or adversarial training, most models still maintain an error rate over 70 percent on DAmageNet. DAmageNet is the first universal adversarial dataset. It could be downloaded freely and serve as a benchmark for robustness testing and adversarial training. [ABSTRACT FROM AUTHOR]

Recently, many researches on convolution neural network (CNN) based in-loop filters have been proposed to improve coding efficiency. However, most existing CNN based filters tend to train and deploy multiple networks for various quantization parameters (QP) and frame types (FT), which drastically increases resources in training these models and the memory burdens for video codec. In this paper, we propose a novel variable CNN (VCNN) based in-loop filter for VVC, which can effectively handle the compressed videos with different QPs and FTs via a single model. Specifically, an efficient and flexible attention module is developed to recalibrate features according to QPs or FTs. Then we embed the module into the residual block so that these informative features can be continuously utilized in the residual learning process. To minimize the information loss in the learning process of the entire network, we utilize a residual feature aggregation module (RFA) for more efficient feature extraction. Based on it, an efficient network architecture VCNN is designed that can not only effectively reduce compression artifacts, but also can be adaptive to various QPs and FTs. To address training data imbalance on various QPs and FTs and improve the robustness of the model, a focal mean square error loss function is employed to train the proposed network. Then we integrate the VCNN into VVC as an additional tool of in-loop filters after the deblocking filter. Extensive experimental results show that our VCNN approach obtains on average 3.63%, 4.36%, 4.23%, 3.56% under all intra, low-delay P, low-delay, and random access configurations, respectively, which is even better than QP-Separate models. [ABSTRACT FROM AUTHOR]

Visual grounding (VG) aims to locate the most relevant object or region in an image, based on a natural language query. Generally, it requires the machine to first understand the query, identify the key concepts in the image, and then locate the target object by specifying its bounding box. However, in many real-world visual grounding applications, we have to face with ambiguous queries and images with complicated scene structures. Identifying the target based on highly redundant and correlated information can be very challenging, and often leading to unsatisfactory performance. To tackle this, in this paper, we exploit an attention module for each kind of information to reduce internal redundancies. We then propose an accumulated attention (A-ATT) mechanism to reason among all the attention modules jointly. In this way, the relation among different kinds of information can be explicitly captured. Moreover, to improve the performance and robustness of our VG models, we additionally introduce some noises into the training procedure to bridge the distribution gap between the human-labeled training data and the real-world poor quality data. With this “noised” training strategy, we can further learn a bounding box regressor, which can be used to refine the bounding box of the target object. We evaluate the proposed methods on four popular datasets (namely ReferCOCO, ReferCOCO+, ReferCOCOg, and GuessWhat?!). The experimental results show that our methods significantly outperform all previous works on every dataset in terms of accuracy. [ABSTRACT FROM AUTHOR]

Keypoints matching is a pivotal component for many image-relevant applications such as image stitching, visual simultaneous localization and mapping (SLAM), and so on. Both handcrafted-based and recently emerged deep learning-based keypoints matching methods merely rely on keypoints and local features, while losing sight of other available sensors such as inertial measurement unit (IMU) in the above applications. In this paper, we demonstrate that the motion estimation from IMU integration can be used to exploit the spatial distribution prior of keypoints between images. To this end, a probabilistic perspective of attention formulation is proposed to integrate the spatial distribution prior into the attentional graph neural network naturally. With the assistance of spatial distribution prior, the effort of the network for modeling the hidden features can be reduced. Furthermore, we present a projection loss for the proposed keypoints matching network, which gives a smooth edge between matching and un-matching keypoints. Image matching experiments on visual SLAM datasets indicate the effectiveness and efficiency of the presented method. [ABSTRACT FROM AUTHOR]

A number of visual tracking methods achieve the state-of-the-art performance based on deep learning recently. However, most of these trackers utilize the deep neural network in regression task or classification task separately. In this paper, we propose an adversarial deep tracking framework. The framework is composed of a fully convolutional Siamese neural network (regression network) and a discriminative classification network. Then, we jointly optimize the regression network and the classification network by adversarial learning. In the uniform framework, the regression network and classification network can be trained end-to-end as a whole using large amounts of video training data sets. During the testing phase, the regression network generates a response map which reflects the location and the size of the target within each candidate search patch, and the classification network discriminates which response map is the best in terms of the corresponding template patch and candidate search patch. In addition, we propose an attention visualization algorithm for our tracker, and it reflects the area that attracts the attention of our tracker during tracking. The experimental results on three large-scale visual tracking benchmarks (OTB-100, TC-128, and VOT2016) demonstrate the effectiveness of the proposed tracking algorithm and show that our tracker performs comparably against the state-of-the-art trackers. [ABSTRACT FROM AUTHOR]

Deep learning can provide rapid brain age estimation based on brain magnetic resonance imaging (MRI). However, most studies use one neural network to extract the global information from the whole input image, ignoring the local fine-grained details. In this paper, we propose a global-local transformer, which consists of a global-pathway to extract the global-context information from the whole input image and a local-pathway to extract the local fine-grained details from local patches. The fine-grained information from the local patches are fused with the global-context information by the attention mechanism, inspired by the transformer, to estimate the brain age. We evaluate the proposed method on 8 public datasets with 8,379 healthy brain MRIs with the age range of 0–97 years. 6 datasets are used for cross-validation and 2 datasets are used for evaluating the generality. Comparing with other state-of-the-art methods, the proposed global-local transformer reduces the mean absolute error of the estimated ages to 2.70 years and increases the correlation coefficient of the estimated age and the chronological age to 0.9853. In addition, our proposed method provides regional information of which local patches are most informative for brain age estimation. Our source code is available on: https://github.com/shengfly/global-local-transformer. [ABSTRACT FROM AUTHOR]

Driving a car in an urban setting is an extremely difficult problem, incorporating a large number of complex visual tasks; however, this problem is solved daily by most adults with little apparent effort. This paper proposes a novel vision-based approach to autonomous driving that can predict and even anticipate a driver's behavior in real time, using preattentive vision only. Experiments on three large datasets totaling over 200 000 frames show that our preattentive model can 1) detect a wide range of driving-critical context such as crossroads, city center, and road type; however, more surprisingly, it can 2) detect the driver's actions (over 80% of braking and turning actions) and 3) estimate the driver's steering angle accurately. Additionally, our model is consistent with human data: First, the best steering prediction is obtained for a perception to action delay consistent with psychological experiments. Importantly, this prediction can be made before the driver's action. Second, the regions of the visual field used by the computational model strongly correlate with the driver's gaze locations, significantly outperforming many saliency measures and comparable to state-of-the-art approaches. [ABSTRACT FROM PUBLISHER]

Channel selection or electrode placement for neural decoding is a commonly encountered problem in electroencephalography (EEG). Since evaluating all possible channel combinations is usually infeasible, one usually has to settle for heuristic methods or convex approximations without optimality guarantees. To date, it remains unclear how large the gap is between the selection made by these approximate methods and the truly optimal selection. The goal of this paper is to quantify this optimality gap for several state-of-the-art channel selection methods in the context of least-squares based neural decoding. To this end, we reformulate the channel selection problem as a mixed-integer quadratic program (MIQP), which allows the use of efficient MIQP solvers to find the optimal channel combination in a feasible computation time for up to 100 candidate channels. As this reveals the exact solution to the combinatorial problem, it allows to quantify the performance losses when using state-of-the-art sub-optimal (yet faster) channel selection methods. In a context of auditory attention decoding, we find that a greedy channel selection based on the utility metric does not show a significant optimality gap compared to optimal channel selection, whereas other state-of-the-art greedy or l 1 -norm penalized methods do show a significant loss in performance. Furthermore, we demonstrate that the MIQP formulation also provides a natural way to incorporate topology constraints in the selection, e.g., for electrode placement in neuro-sensor networks with galvanic separation constraints. Furthermore, a combination of this utility-based greedy selection with an MIQP solver allows to perform a topology constrained electrode placement, even in large scale problems with more than 100 candidate positions.

Pancreatic ductal adenocarcinoma (PDAC) is the third most common cause of cancer death in the United States. Predicting tumors like PDACs (including both classification and segmentation) from medical images by deep learning is becoming a growing trend, but usually a large number of annotated data are required for training, which is very labor-intensive and time-consuming. In this paper, we consider a partially supervised setting, where cheap image-level annotations are provided for all the training data, and the costly per-voxel annotations are only available for a subset of them. We propose an Inductive Attention Guidance Network (IAG-Net) to jointly learn a global image-level classifier for normal/PDAC classification and a local voxel-level classifier for semi-supervised PDAC segmentation. We instantiate both the global and the local classifiers by multiple instance learning (MIL), where the attention guidance, indicating roughly where the PDAC regions are, is the key to bridging them: For global MIL based normal/PDAC classification, attention serves as a weight for each instance (voxel) during MIL pooling, which eliminates the distraction from the background; For local MIL based semi-supervised PDAC segmentation, the attention guidance is inductive, which not only provides bag-level pseudo-labels to training data without per-voxel annotations for MIL training, but also acts as a proxy of an instance-level classifier. Experimental results show that our IAG-Net boosts PDAC segmentation accuracy by more than 5% compared with the state-of-the-arts. [ABSTRACT FROM AUTHOR]

In the task of person re-identification, the attention mechanism and fine-grained information have been proved to be effective. However, it has been observed that models often focus on the extraction of features with strong discrimination, and neglect other valuable features. The extracted fine-grained information may include redundancies. In addition, current methods lack an effective scheme to remove background interference. Therefore, this paper proposes the feature refinement and filter network to solve the above problems from three aspects: first, by weakening the high response features, we aim to identify highly valuable features and extract the complete features of persons, thereby enhancing the robustness of the model; second, by positioning and intercepting the high response areas of persons, we eliminate the interference arising from background information and strengthen the response of the model to the complete features of persons; finally, valuable fine-grained features are selected using a multi-branch attention network for person re-identification to enhance the performance of the model. Our extensive experiments on the benchmark Market-1501, DukeMTMC-reID, CUHK03 and MSMT17 person re-identification datasets demonstrate that the performance of our method is comparable to that of state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Accurate segmentation of the prostate is a key step in external beam radiation therapy treatments. In this paper, we tackle the challenging task of prostate segmentation in CT images by a two-stage network with 1) the first stage to fast localize, and 2) the second stage to accurately segment the prostate. To precisely segment the prostate in the second stage, we formulate prostate segmentation into a multi-task learning framework, which includes a main task to segment the prostate, and an auxiliary task to delineate the prostate boundary. Here, the second task is applied to provide additional guidance of unclear prostate boundary in CT images. Besides, the conventional multi-task deep networks typically share most of the parameters (i.e., feature representations) across all tasks, which may limit their data fitting ability, as the specificity of different tasks are inevitably ignored. By contrast, we solve them by a hierarchically-fused U-Net structure, namely HF-UNet. The HF-UNet has two complementary branches for two tasks, with the novel proposed attention-based task consistency learning block to communicate at each level between the two decoding branches. Therefore, HF-UNet endows the ability to learn hierarchically the shared representations for different tasks, and preserve the specificity of learned representations for different tasks simultaneously. We did extensive evaluations of the proposed method on a large planning CT image dataset and a benchmark prostate zonal dataset. The experimental results show HF-UNet outperforms the conventional multi-task network architectures and the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

An EEG-based Brain-Computer Interface (BCI) is a system that enables a user to communicate with and intuitively control external devices solely using the user's intentions. Current EEG-based BCI research usually involves a subject-specific adaptation step before a BCI system is ready to be employed by a new user. However, the subject-independent scenario, in which a well-trained model can be directly applied to new users without pre-calibration, is particularly desirable yet rarely explored. Considering this critical gap, our focus in this paper is the subject-independent scenario of EEG-based human intention recognition. We present a G raph-based H ierarchical A ttention M odel (G-HAM) that utilizes the graph structure to represent the spatial information of EEG sensors and the hierarchical attention mechanism to focus on both the most discriminative temporal periods and EEG nodes. Extensive experiments on a large EEG dataset containing 105 subjects indicate that our model is capable of exploiting the underlying invariant EEG patterns across different subjects and generalizing the patterns to new subjects with better performance than a series of state-of-the-art and baseline approaches.

Human action recognition in 3D skeleton sequences has attracted a lot of research attention. Recently, long short-term memory (LSTM) networks have shown promising performance in this task due to their strengths in modeling the dependencies and dynamics in sequential data. As not all skeletal joints are informative for action recognition, and the irrelevant joints often bring noise which can degrade the performance, we need to pay more attention to the informative ones. However, the original LSTM network does not have explicit attention ability. In this paper, we propose a new class of LSTM network, global context-aware attention LSTM, for skeleton-based action recognition, which is capable of selectively focusing on the informative joints in each frame by using a global context memory cell. To further improve the attention capability, we also introduce a recurrent attention mechanism, with which the attention performance of our network can be enhanced progressively. Besides, a two-stream framework, which leverages coarse-grained attention and fine-grained attention, is also introduced. The proposed method achieves state-of-the-art performance on five challenging datasets for skeleton-based action recognition. [ABSTRACT FROM PUBLISHER]

Driver vigilance estimation is an important task for transportation safety. Wearable and portable brain-computer interface devices provide a powerful means for real-time monitoring of the vigilance level of drivers to help with avoiding distracted or impaired driving. In this paper, we propose a novel multimodal architecture for in-vehicle vigilance estimation from Electroencephalogram and Electrooculogram. To enable the system to focus on the most salient parts of the learned multimodal representations, we propose an architecture composed of a capsule attention mechanism following a deep Long Short-Term Memory (LSTM) network. Our model learns hierarchical dependencies in the data through the LSTM and capsule feature representation layers. To better explore the discriminative ability of the learned representations, we study the effect of the proposed capsule attention mechanism including the number of dynamic routing iterations as well as other parameters. Experiments show the robustness of our method by outperforming other solutions and baseline techniques, setting a new state-of-the-art. We then provide an analysis on different frequency bands and brain regions to evaluate their suitability for driver vigilance estimation. Lastly, an analysis on the role of capsule attention, multimodality, and robustness to noise is performed, highlighting the advantages of our approach. [ABSTRACT FROM AUTHOR]

Inferring object-wise human attention in 3D space from the third-person perspective (e.g., a camera) is crucial to many visual tasks and applications, including human-robot collaboration, unmanned vehicle driving, etc. Challenges arise from classical human attention when human eyes are not visible to cameras, gaze point is outside the field of vision, or the gazed object is occluded by others in the 3D space. In this case, blind 3D human attention inference brings a new paradigm to the community. In this paper, we address these challenges by proposing a scene-behavior associated mechanism, in which both 3D scene and temporal behavior of human are adopted to infer object-wise human attention and its transition. Specifically, point cloud is reconstructed and used for the spatial representation of 3D scene, which is beneficial to handle the blind problem from the perspective of a camera. Based on this, in order to address the blind human attention inference without eye information, we propose a Sequential Skeleton Based Attention Network (S2BAN) for behavior-based attention modeling. As is embedded in the scene-behavior associated mechanism, the proposed S2BAN is built under the temporal architecture of Long-Short-Term-Memory (LSTM). Our network employs human skeleton as behavior representation, and maps it to the attention direction frame by frame, which makes attention inference a temporal-correlated issue. With the help of S2BAN, 3D gaze spot and further the attended objects can be obtained frame by frame via intersection and segmentation on the previously reconstructed point cloud. Finally, we conduct experiments from various aspects to verify the object-wise attention localization accuracy, the angular error of attention direction calculation, as well as the subjective results. The experimental results show that the proposed outperforms other competitors. [ABSTRACT FROM AUTHOR]

Image co-segmentation is an active computer vision task that aims to segment the common objects from a set of images. Recently, researchers design various learning-based algorithms to undertake the co-segmentation task. The main difficulty in this task is how to effectively transfer information between images to make conditional predictions. In this paper, we present CycleSegNet, a novel framework for the co-segmentation task. Our network design has two key components: a region correspondence module which is the basic operation for exchanging information between local image regions, and a cycle refinement module, which utilizes ConvLSTMs to progressively update image representations and exchange information in a cycle and iterative manner. Extensive experiments demonstrate that our proposed method significantly outperforms the state-of-the-art methods on four popular benchmark datasets — PASCAL VOC dataset, MSRC dataset, Internet dataset, and iCoseg dataset, by 2.6%, 7.7%, 2.2%, and 2.9%, respectively. [ABSTRACT FROM AUTHOR]

Defocus blur detection (DBD), which has been widely applied to various fields, aims to detect the out-of-focus or in-focus pixels from a single image. Despite the fact that the deep learning based methods applied to DBD have outperformed the hand-crafted feature based methods, the performance cannot still meet our requirement. In this paper, a novel network is established for DBD. Unlike existing methods which only learn the projection from the in-focus part to the ground-truth, both in-focus and out-of-focus pixels, which are completely and symmetrically complementary, are taken into account. Specifically, two symmetric branches are designed to jointly estimate the probability of focus and defocus pixels, respectively. Due to their complementary constraint, each layer in a branch is affected by an attention obtained from another branch, effectively learning the detailed information which may be ignored in one branch. The feature maps from these two branches are then passed through a unique fusion block to simultaneously get the two-channel output measured by a complementary loss. Additionally, instead of estimating only one binary map from a specific layer, each layer is encouraged to estimate the ground truth to guide the binary map estimation in its linked shallower layer followed by a top-to-bottom combination strategy, gradually exploiting the global and local information. Experimental results on released datasets demonstrate that our proposed method remarkably outperforms state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Person re-identification (Re-ID) aims to match pedestrian images across various scenes in video surveillance. There are a few works using attribute information to boost Re-ID performance. Specifically, those methods leverage attribute information to boost Re-ID performance by introducing auxiliary tasks like verifying the image level attribute information of two pedestrian images or recognizing identity level attributes. Identity level attribute annotations cost less manpower and are well-fitted for person re-identification task compared with image-level attribute annotations. However, the identity attribute information may be very noisy due to incorrect attribute annotation or lack of discriminativeness to distinguish different persons, which is probably unhelpful for the Re-ID task. In this paper, we propose a novel Attribute Attentional Block (AAB), which can be integrated into any backbone network or framework. Our AAB adopts reinforcement learning to drop noisy attributes based on our designed reward and then utilizes aggregated attribute attention of the remaining attributes to facilitate the Re-ID task. Experimental results demonstrate that our proposed method achieves state-of-the-art results on three benchmark datasets. [ABSTRACT FROM AUTHOR]

Recently, automatic diagnostic approaches have been widely used to classify ocular diseases. Most of these approaches are based on a single imaging modality (e.g., fundus photography or optical coherence tomography (OCT)), which usually only reflect the oculopathy to a certain extent, and neglect the modality-specific information among different imaging modalities. This paper proposes a novel modality-specific attention network (MSAN) for multi-modal retinal image classification, which can effectively utilize the modality-specific diagnostic features from fundus and OCT images. The MSAN comprises two attention modules to extract the modality-specific features from fundus and OCT images, respectively. Specifically, for the fundus image, ophthalmologists need to observe local and global pathologies at multiple scales (e.g., from microaneurysms at the micrometer level, optic disc at millimeter level to blood vessels through the whole eye). Therefore, we propose a multi-scale attention module to extract both the local and global features from fundus images. Moreover, large background regions exist in the OCT image, which is meaningless for diagnosis. Thus, a region-guided attention module is proposed to encode the retinal layer-related features and ignore the background in OCT images. Finally, we fuse the modality-specific features to form a multi-modal feature and train the multi-modal retinal image classification network. The fusion of modality-specific features allows the model to combine the advantages of fundus and OCT modality for a more accurate diagnosis. Experimental results on a clinically acquired multi-modal retinal image (fundus and OCT) dataset demonstrate that our MSAN outperforms other well-known single-modal and multi-modal retinal image classification methods. [ABSTRACT FROM AUTHOR]

In this paper, we address a problem of view-disturbing raindrop removal on a single image. In existing methods to tackle this problem, machine learning based ones seem promising but require elaborate pairwise images, i.e., the raindrop-degraded image and the corresponding clean image of the same scene, for training. To overcome this drawback, we propose a weakly supervised learning based model in the absence of pairwise training examples, which needs only a collection of images with image-level annotations indicating the presence/absence of raindrops for training. Specifically, we train a raindrop detector for highlighting regions of raindrops in a multi-task learning manner. Then, we propose an attention-based generative network for raindrop removal and introduce a weighted preservation loss to retain the non-raindrop details. Specially, our model can be mixedly trained with pairwise and unpaired samples, which enables us to conveniently adapt the model to a new domain. Experiments verify the effect of the proposed method. Especially, using only weakly-supervised learning, our method can achieve comparable results with state-of-the-art strongly-supervised learning methods. [ABSTRACT FROM AUTHOR]

In the context of the hierarchy of risk controls, warnings, administrative controls, and personal protective equipment are sometimes referred to as lower order controls. In practice, the effectiveness of these controls can be overestimated due to a misunderstanding of the human factors that affect error likelihood in their implementation. This paper explores one human factor: the limitation of attention on residual risk inherent in lower order controls. The limitation of attention has implications for identifying hazards, assessing risk, taking appropriate actions based on warnings, executing safe work practices, and proper selection and use of personal protective equipment. Raising awareness and improved understanding of these topics can enable more effective risk assessment, quantification of residual risk, and more effective selection and combination of risk controls to achieve risk as low as reasonably practicable. [ABSTRACT FROM PUBLISHER]

This paper comprehensively reviews medical VQA models, structures, and datasets, focusing on combining vision and language. Over 75 models and their statistical and SWOT (Strengths, Weaknesses, Opportunities, Threats) analyses were compared and analyzed. The study highlights whether the researchers in the general field influence those in the medical field. According to an analysis of text encoding techniques, LSTM is the approach that is utilized the most (42%), followed by non-text methods (14%) and BiLSTM (12%), whereas VGGNet (40%) and ResNet (22%) are the most often used vision methods, followed by Ensemble approaches (16%). Regarding fusion techniques, 14% of the models employed non-specific methods, while SAN (13%) and concatenation (10%) were frequently used. The study identifies LSTM-VGGNet and LSTM-ResNet combinations as the primary approaches in medical VQA, with 18% and 15% usage rates, respectively. The statistical analysis of medical VQA from 2018 to 2023 and individual yearly analyses reveals consistent preferences for LSTM and VGGNet, except in 2018 when ResNet was more commonly used. The SWOT analysis provides insights into the strengths and weaknesses of medical VQA research, highlighting areas for future exploration. These areas include addressing limited dataset sizes, enhancing question diversity, mitigating unimodal bias, exploring multi-modal datasets, leveraging external knowledge, incorporating multiple images, ensuring practical medical application integrity, improving model interpretation, and refining evaluation methods. This paper’s findings contribute to understanding medical VQA and offer valuable guidance for future researchers aiming to make advancements in this field.

Object detection based on unmanned aerial vehicle (UAV) platforms is essential for both engineering and research. Complex scale problems in UAV application scenarios require strong regression localization capabilities from target detection algorithms. Nonetheless, due to the constraints of UAV platform, it is difficult to increase accuracy by deepening the network. Therefore, this paper presents an improved YOLOv5 with an attention mechanism, consisting a Convolution-Swin Transformer Block (CSTB) utilizing Swin Transformer as well as a Convolution-block Attention Module (CBAM) to improve network positioning accuracy. In addition, this paper incorporates Bidirectional Feature Pyramid Network (BiFPN), Spatial Pyramid Pooling-Fast (SPPF) and some network components to increase the average precision while maintaining the limited size of the model. Experiments on Visdrone2019 dataset show that the proposed approach can raise the mean Average Precision (mAP) by 5.4% compared to YOLOv5, with only 18% increase in model size.

Recognition of floor plans has been a challenging and popular task. However, existing approaches are struggling to make accurate room-level unified predictions, seriously limiting their visual quality and applicability. In this paper, we propose a novel approach to recognize the floor plan layouts with a newly proposed Offset-Guided Attention mechanism to improve the semantic consistency within a room. In addition, we present a Feature Fusion Attention module that leverages the channel-wise attention to encourage the consistency of the room, wall, and door predictions, further enhancing the room-level semantic consistency. Experimental results manifest our approach is able to improve room-level semantic consistency and outperforms the existing works both qualitatively and quantitatively. To summarize, this paper tackles inconsistent semantic prediction issues in floor plan segmentation and enhances visual plausibility, quantitative performance, and practical applicability.

Modern deep learning algorithms comprise highly complex artificial neural networks, making it extremely difficult for humans to track their inference processes. As the social implementation of deep learning progresses, the human and economic losses caused by inference errors are becoming increasingly problematic, making it necessary to develop methods to explain the basis for the decisions of deep learning algorithms. Although an attention mechanism-based method to visualize the regions that contribute to steering angle prediction in an automated driving task has been proposed, its explanatory capability is low. In this paper, we focus on the fact that the importance of each bit in the activation value of a network is biased (i.e., the sign and exponent bits are weighted more heavily than the mantissa bits), which has been overlooked in previous studies. Specifically, this paper quantizes network activations, encouraging important information to be aggregated to the sign bit. Further, we introduce an attention mechanism restricted to the sign bit to improve the explanatory power. Our numerical experiment using the Udacity dataset revealed that the proposed method achieves a $1.14\times $ higher area under curve (AUC) in terms of the deletion metric.

In this paper, we propose a human pose estimation algorithm for an impulse radio ultra-wideband (IR-UWB) radar based on the transformer-based deep learning model. We have built an IR-UWB radar system with an 8-by-8 multiple-input multiple-output (MIMO) antenna array. The IR-UWB radar system in our paper is advantageous for the through-wall detection application since it operates on a very low frequency range (i.e., 0.45 to 3.55 GHz) compared to other existing works on RF-based human pose estimation. Moreover, the human pose estimation by an IR-UWB radar has not been studied very well in other existing works since all existing works have used a frequency-modulated continuous wave (FMCW) radar or a WiFi device. We propose a 3D-TransPOSE algorithm for the 3D human pose estimation from the IR-UWB radar signals. The proposed algorithm is designed based on the transformer architecture. While the transformer has actively been studied in the natural language processing (NLP) or vision domains, no prior work has applied the transformer model to the RF-based human pose estimation problem. The attention mechanism of the proposed algorithm is able to focus on the relevant time segments of the IR-UWB radar signals for detecting the human pose, eliminating the needs of converting radar signals to a voxelized 3D image. We have gathered a large dataset of IR-UWB radar signals labeled with 3D human skeletons, and shown that the proposed algorithm is able to detect human skeletons with a high accuracy.