The surface electromyography (sEMG)-based gesture recognition with deep learning approach plays an increasingly important role in human-computer interaction. Existing architectures are mainly based on Convolutional Neural Network (CNN) architecture which captures spatial information of electromyogram signal. Motivated by the sequential nature signal, we propose attention-based hybrid CNN and RNN (CNN-RNN) to better capture temporal properties signal for problem. Moreover, present a new sEMG...

Gesture recognition using sparse multichannel surface electromyography (sEMG) is a challenging problem, and the solutions are far from optimal point of view muscle-computer interface. In this paper, we address problem context multi-view deep learning. A novel convolutional neural network (CNN) framework proposed by combining classical sEMG feature sets with CNN-based learning model. The consists two parts. first part, representations modeled in parallel multistream CNN, performance-based...

We propose an experimental scheme to implement a strong photon blockade with single quantum dot coupled nanocavity. The effect can be tremendously enhanced by driving the cavity and simultaneously two classical laser fields. This enhancement of is ascribed interference avoid two-photon excitation field. Comparing Jaynes-Cummings model, second-order correlation function at zero time delay $g^{(2)}(0)$ in our reduced orders magnitude system sustains large intracavity number. A red (blue)...

The creation of improvised dancing choreographies is an important research field cross-modal analysis. A key point this task how to effectively create and correlate music dance with a probabilistic one-to-many mapping, which essential realistic dances various genres. To address issue, we propose GAN-based association framework, DeepDance, correlates two different modalities (dance motion music) together, aiming at creating the desired sequence in terms input music. Its generator predictively...

We introduce a novel method for synthesizing dance motions that follow the emotions and contents of piece music. Our employs learning-based approach to model music motion mapping relationship embodied in example along with those motions' accompanying background A key step our is train matching quality rating function through learning exhibited synchronized data, which were captured from professional human performance. To generate an optimal sequence segments match music, we constraint-based...

Fine-grained grocery product recognition via camera is a challenging task to identify the visually similar products with subtle differences by using single-shot training examples. To address this issue? we present novel hybrid classification approach that combines feature-based matching and one-shot deep learning coarse-to-fine strategy. The candidate regions of instances are first detected coarsely labeled recurring features in images without any training. Then, attention maps generated...

Conventionally, gesture recognition based on non-intrusive muscle-computer interfaces required a strongly-supervised learning algorithm and large amount of labeled training signals surface electromyography (sEMG). In this work, we show that temporal relationship sEMG data glove provides implicit supervisory signal for the model. To demonstrate this, present semi-supervised framework with novel Siamese architecture sEMG-based recognition. Specifically, employ auxiliary tasks to learn visual...

Traditional evaluation method of camouflage texture effect is subjective evaluation. It's very tedious and inconvenient to direct the designing. In this paper, a systemic rational for direction designing proposed. A new based on WSSIM (Weight structural similarity) given access effects at first. Then nature image features between background are calculated help texture. Primary experimental results show that proposed helpful design

3D pose estimation has attracted increasing attention with the availability of high-quality benchmark datasets. However, prior works show that deep learning models tend to learn spurious correlations, which fail generalize beyond specific dataset they are trained on. In this work, we take a step towards training robust for cross-domain task, brings together ideas from causal representation and generative adversarial networks. Specifically, paper introduces novel framework explicitly exploits...

The conventional surface electromyography (sEMG)-based gesture recognition systems exhibit impressive performance in controlled laboratory settings. As most are trained a closed-set setting, the systems's may see significant deterioration when novel gestures presented as imposter. In addition, state-of-the-art generative and discriminative methods have achieved considerable on high-density sEMG signals. This can be seen an unrealistic setting real-world muscle computer interface mainly...

To improve the accuracy of surface electromyography (sEMG)-based gesture recognition, we present a novel hybrid approach that combines real sEMG signals with corresponding virtual hand poses. The poses are generated by means proposed cross-modal association model constructed based on adversarial learning to capture intrinsic relationship between and We report comprehensive evaluations for both frame- window-based recognitions seven-sparse-multichannel four-high-density-benchmark databases....

Sensor-based human activity recognition aims at detecting various physical activities performed by people with ubiquitous sensors. Different from existing deep learning-based method which mainly extracting black-box features the raw sensor data, we propose a hierarchical multi-view aggregation network based on feature spaces. Specifically, first construct views of spaces for each individual in terms white-box and features. Then our model learns unified representation aggregating context...

Electronic skin (e-skin) with excellent flexibility and comfortable wearability has attracted considerable attention in the past decade. Flexible strain sensors are one of key elements e-skins, which always attached to joints humans or robots for motion monitoring. However, most previously reported cannot fit different measure bending angles without calibration. This work aims propose a flexible sensor an adaptive model automatically calculating angle joints. The proposed is screen-printed...

Affective computation generally focuses on the informatics of affect: structuring, formalizing, and representing emotion as informational units. We propose instead an enigmatics affect, a critical technical practice that respects rich undefinable complexities human affective experience. Our interactive installation, Influencing Machine, allows users to explore dynamic landscape emotionally expressive sound child-like drawings, using tangible, intuitive input device supports open-ended...

Training an accurate 3D human pose estimator often requires a large amount of ground-truth data which is inefficient and costly to collect. Previous methods have either resorted weakly supervised reduce the demand for training, or using synthetically-generated but photo-realistic samples enlarge training pool. Nevertheless, former mainly require additional supervision, such as unpaired data, camera parameters in multiview settings. On other hand, latter accurately textured models,...

The two-dimensional (2D) and three-dimensional (3D) user interfaces have been a prominent topic of augmented reality (AR) research, their impact on the efficacy usability one-time tasks has extensively examined. As AR is increasingly adopted in industry for repetitive tasks, there an urgent need research into effect 2D 3D interfaces. In this study, we developed two prototypes conducted comparison study with forty participants to assess respective influence cognitive load, perceived...

Sketch-drawings is an intuitive and comprehensive means of conveying movement ideas in character animation. We proposed a novel sketch-based approach to assisting the authoring choreographing Kungfu motions at early stage animation creation. Given two human figure sketches corresponding initial closing posture form, trajectory drawings on specific moving joints, MotionMaster can directly rapid-prototype realistic 3D motion sequence by retrieval refinement based database. The animators then...

Speech-driven gesture generation is an emerging field within virtual human creation. However, a significant challenge lies in accurately determining and processing the multitude of input features (such as acoustic, semantic, emotional, personality, even subtle unknown features). Traditional approaches, reliant on various explicit feature inputs complex multimodal processing, constrain expressiveness resulting gestures limit their applicability. To address these challenges, we present...

Conditional human motion synthesis (HMS) aims to generate sequences that conform specific conditions. Text and audio represent the two predominant modalities employed as HMS control While existing research has primarily focused on single conditions, multi-condition remains underexplored. In this study, we propose a framework, termed MCM, based dual-branch structure composed of main branch branch. This framework effectively extends applicability diffusion model, which is initially predicated...

The recent progress in sEMG-based hand gesture detection has developed a set of predefined gestures for interaction. However, customized are less concerned due to the lack supporting tools training alternative gestures. To fill gap, we present system, called CAPG-MYO, user-defined An armband named CAPG was used simultaneously capture sEMG and IMU signals from participants construct small-scale dataset with predict gestures, multiview convolutional neural network handle consequently shaped...