In this paper, we present a method named Cross-Modal Knowledge Adaptation (CMKA) for language-based person search. We argue that the image and text information are not equally important in determining person's identity. other words, carries image-specific such as lighting condition background, while contains more modal agnostic is beneficial to cross-modal matching. Based on consideration, propose CMKA adapt knowledge of text. Specially, text-to-image guidance obtained at different levels:...

The electroencephalography classifier is the most important component of brain-computer interface based systems. There are two major problems hindering improvement it. First, traditional methods do not fully exploit multimodal information. Second, large-scale annotated EEG datasets almost impossible to acquire because biological data acquisition challenging and quality annotation costly. Herein, we propose a novel deep transfer learning approach solve these problems. model cognitive events...

Taking the feature pyramids into account has become a crucial way to boost object detection performance. While various pyramid representations have been developed, previous works are still inefficient integrate semantical information over different scales. Moreover, recent detectors suffering from accurate location applications, mainly due coarse definition of positive examples at training and predicting phases. In this paper, we begin by analyzing current solutions, then propose novel...

In this work, we propose a division-and-summarization (DaS) framework for dense video captioning. After partitioning each untrimmed long as multiple event proposals, where proposal consists of set short segments, extract visual feature (e.g., C3D feature) from segment and use the existing image/video captioning approach to generate one sentence description segment. Considering that generated sentences contain rich semantic descriptions about whole proposal, formulate task cue aided...

As a new classification platform, deep learning has recently received increasing attention from researchers and been successfully applied to many domains. In some domains, like bioinformatics robotics, it is very difficult construct large-scale well-annotated dataset due the expense of data acquisition costly annotation, which limits its development. Transfer relaxes hypothesis that training must be independent identically distributed (i.i.d.) with test data, motivates us use transfer solve...

The brain–computer interface-based rehabilitation robot has quickly become a very important research area due to its natural interaction. One of the most problems in interface is that large-scale annotated electroencephalography data sets required by advanced classifiers are almost impossible acquire because biological acquisition challenging and quality annotation costly. Transfer learning relaxes hypothesis training must be independent identically distributed with test data. It can...

The control of a high Degree Freedom (DoF) robot to grasp target in three-dimensional space using Brain-Computer Interface (BCI) remains very difficult problem solve. Design synchronous BCI requires the user perform brain activity task all time according predefined paradigm; such process is boring and fatiguing. Furthermore, strategy switching between robotic auto-control not reliable because accuracy Motor Imagery (MI) pattern recognition rarely reaches 100%. In this paper, an asynchronous...

It is hard to grasp objects based on brain-computer interface (BCI) by brain-actuated robot arm and hand due its high degree of freedom. Shared control strategy hybrid BCI are research trends solve this problem brainactuated discrete event system. We propose a new shared method fused fuzzy Petri nets (PNs) for combining the automatic (AC) control. This takes advantages both PNs such as easy modeling, robustness, effectiveness. Both MATLAB simulation test Barrett practical experiments show...

Tactile sensing is becoming an indispensable robotic ability for object recognition and grasping manipulation despite dealing with tactile data as the force distribution over array sensors continuously changes a function of time. In this paper, we propose efficient feature extractor named linear dynamic systems based fuzzy C-means method (LDS) to encode sequences, both spatially temporally. To end, decompose every input sequence into multiple subsequences, each which locally described by...

Different functional areas of the human brain play different roles in activity, which has not been paid sufficient research attention brain-computer interface (BCI) field. This paper presents a new approach for electroencephalography (EEG) classification that applies attention-based transfer learning. Our considers importance to improve accuracy EEG classification, and provides an additional way automatically identify associated with activities without involvement medical professional. We...

Brain-Computer Interfaces (BCI) can help disable people to improve human — environment interaction and rehabilitation. Grasping objects with EEG-based BCI has become a popular hard research in recent years due the high degree of freedom robot complex grasp planning. Unlike commonly used paradigms, we propose pipeline hybrid for grasping by shared control solve key problems including target object selection, intelligent planning both user intension robot. Six experimental users could...

To achieve better results on object detection for autonomous vehicle under complex outdoor conditions, we attempt to integrated the sensor-fusion, hierarchical multi-view networks and traditional heuristical method together. The most significant environmental perception sensors vehicles are camera LIDAR. 2D RGB image 3D point cloud from LIDAR respectively utilized. proposal network (HMVPN) is proposed in this paper, which can effectively fuse multi-modal information of with As there several...

Brain computer interface (BCI) is the only way for some special patients to communicate with outside world and provide a direct control channel between brain external devices. As non-invasive interface, scalp electroencephalography (EEG) has significant potential be major input signal future BCI systems. Traditional methods focus on particular feature in EEG signal, which limits practical applications of EEG-based BCI. In this paper, we propose algorithm classification ability fuse multiple...

The common spatial pattern (CSP) and other spatiospectral feature extraction methods have become the most effective successful approaches to solve problem of motor imagery electroencephalography (MI-EEG) recognition from multichannel neural activity in recent years. However, these need a lot preprocessing postprocessing such as filtering, demean, fusion, which influence classification accuracy easily. In this paper, we utilize linear dynamical systems (LDSs) for EEG signals classification....

In this paper, we first extract three different kinds of high-level features from LIDAR point cloud, and combine them into the DHA (Depth, Height Angle) channels. Integrated with traditional RGB image camera, build a rich feature-based road object classifier by training deep convolutional neural network model six-channel (RGBDHA) data. Subsequently, convolution is fed integration spacial information. With additional upsampled data, reaches higher accuracy than single base methods. Several...

Insufficient training data is a serious problem in all domains related to bioinformatics. Large-scale annotated electroencephalography (EEG) datasets are almost impossible acquire because biological acquisition challenging and quality annotation costly. Transfer learning relaxes the hypothesis that must be independent identically distributed (i.i. d.) with test data, which motivates us use transfer solve of insufficient We propose new approach knowledge via deep framework, includes an...

The brain-actuated robot grasping control is a hard and complex problem due to the low classification accuracy of mental pattern high degree freedom hand arm. We propose an asynchronous MI-based BCI that allows user moving direction arm by intension whenever he feels necessary adjust path. In order improve real-time performance, thresholding method LDS modeling approach are employed for online motor imagery recognition. Then, shared strategy combines automatic our grasp object with obstacle....

Insufficient training data is a serious problem in all domains related to bioinformatics. Transfer learning promising tool solve this problem, which relaxes the hypothesis that must be independent and identically distributed with test data. We construct sophisticated electroencephalography (EEG) signal representation obtain an efficient EEG feature extractor through manifold constraints-based joint adversarial from other domains. more easily distinguished space mapped by extractor. Negative...