A5076090670- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Face recognition and analysis
- Anomaly Detection Techniques and Applications
- Face and Expression Recognition
- Advanced Neural Network Applications
- Gait Recognition and Analysis
- Advanced Image and Video Retrieval Techniques
- Reinforcement Learning in Robotics
- Biometric Identification and Security
- Hand Gesture Recognition Systems
- Artificial Intelligence in Games
- Fire Detection and Safety Systems
- Advanced Vision and Imaging
- Generative Adversarial Networks and Image Synthesis
- Remote-Sensing Image Classification
- Image Enhancement Techniques
- Multimodal Machine Learning Applications
- Visual Attention and Saliency Detection
- Domain Adaptation and Few-Shot Learning
- Video Analysis and Summarization
- Human Motion and Animation
- Evolutionary Game Theory and Cooperation
- Advanced Image Processing Techniques
- Advanced Bandit Algorithms Research
Tsinghua University
2009-2025
University of Chinese Academy of Sciences
2017-2023
Qinghai University
2023
Beijing Academy of Artificial Intelligence
2023
Shandong Institute of Automation
2013-2021
Chinese Academy of Sciences
2013-2021
Institute of Automation
2013-2021
Chongqing Institute of Green and Intelligent Technology
2021
Birkbeck, University of London
2021
Carnegie Mellon University
2020
Siamese network based trackers formulate tracking as convolutional feature cross-correlation between target template and searching region. However, still have accuracy gap compared with state-of-the-art algorithms they cannot take advantage of from deep networks, such ResNet-50 or deeper. In this work we prove the core reason comes lack strict translation invariance. By comprehensive theoretical analysis experimental validations, break restriction through a simple yet effective spatial aware...
Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations the complex view variations exhibited by captured images significantly increase difficulty of learning features from images. To overcome these difficulties, this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature models end end. Our deep architecture explicitly leverages human part cues alleviate robust representations both global...
Skeleton based action recognition distinguishes human actions using the trajectories of skeleton joints, which provide a very good representation for describing actions. Considering that recurrent neural networks (RNNs) with Long Short-Term Memory (LSTM) can learn feature representations and model long-term temporal dependencies automatically, we propose an end-to-end fully connected deep LSTM network recognition. Inspired by observation co-occurrences joints intrinsically characterize...
Human action recognition is an important task in computer vision. Extracting discriminative spatial and temporal features to model the evolutions of different actions plays a key role accomplishing this task. In work, we propose end-to-end attention for human from skeleton data. We build our on top Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM), which learns selectively focus joints within each frame inputs pays levels outputs frames. Furthermore, ensure effective...
The Visual Object Tracking challenge 2015, VOT2015, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. Results 62 are presented. number tested makes VOT 2015 the largest benchmark on tracking to date. For each participating tracker, a short description is provided in appendix. Features VOT2015 go beyond its VOT2014 predecessor are: (i) new dataset twice as large with full annotation targets by rotated bounding boxes and...
Skeleton-based human action recognition has recently attracted increasing attention due to the popularity of 3D skeleton data. One main challenge lies in large view variations captured actions. We propose a novel adaptation scheme automatically regulate observation viewpoints during occurrence an action. Rather than re-positioning skeletons based on defined prior criterion, we design adaptive recurrent neural network (RNN) with LSTM architecture, which enables itself adapt most suitable from...
Offline training for object tracking has recently shown great potentials in balancing accuracy and speed. However, it is still difficult to adapt an offline trained model a target tracked online. This work presents Residual Attentional Siamese Network (RASNet) high performance tracking. The RASNet reformulates the correlation filter within framework, introduces different kinds of attention mechanisms without updating In particular, by exploiting general attention, adapted residual channel...
Skeleton-based human action recognition has attracted great interest thanks to the easy accessibility of skeleton data. Recently, there is a trend using very deep feedforward neural networks model 3D coordinates joints without considering computational efficiency. In this paper, we propose simple yet effective semantics-guided network (SGN) for skeleton-based recognition. We explicitly introduce high level semantics (joint type and frame index) into enhance feature representation capability....
The Visual Object Tracking challenge VOT2017 is the fifth annual tracker benchmarking activity organized by VOT initiative. Results of 51 trackers are presented; many state-of-the-art published at major computer vision conferences or journals in recent years. evaluation included standard and other popular methodologies a new "real-time" experiment simulating situation where processes images as if provided continuously running sensor. Performance tested typically far exceeds baselines. source...
Human action recognition is an important task in computer vision. Extracting discriminative spatial and temporal features to model the evolutions of different actions plays a key role accomplishing this task. In work, we propose end-to-end attention for human from skeleton data. We build our on top Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM), which learns selectively focus joints within each frame inputs pays levels outputs frames. Furthermore, ensure effective...
Skeleton-based human action recognition has recently attracted increasing attention thanks to the accessibility and popularity of 3D skeleton data. One key challenges in lies large variations representations when they are captured from different viewpoints. In order alleviate effects view variations, this paper introduces a novel adaptation scheme, which automatically determines virtual observation viewpoints over course an learning based data driven manner. Instead re-positioning skeletons...
Visual tracking has attracted a significant attention in the last few decades. The recent surge number of publications on tracking-related problems have made it almost impossible to follow developments field. One reasons is that there lack commonly accepted annotated data-sets and standardized evaluation protocols would allow objective comparison different methods. To address this issue, Object Tracking (VOT) workshop was organized conjunction with ICCV2013. Researchers from academia as well...
Discriminant Correlation Filters (DCF) based methods now become a kind of dominant approach to online object tracking. The features used in these methods, however, are either on hand-crafted like HoGs, or convolutional trained independently from other tasks image classification. In this work, we present an end-to-end lightweight network architecture, namely DCFNet, learn the and perform correlation tracking process simultaneously. Specifically, treat DCF as special filter layer added Siamese...
Regression based facial landmark detection methods usually learns a series of regression functions to update the positions from an initial estimation. Most existing approaches focus on learning effective mapping with robust image features improve performance. The approach dealing initialization issue, however, receives relatively fewer attentions. In this paper, we present deep architecture two-stage re-initialization explicitly deal problem. At global stage, given rough face result, full...
Pose variation is one key challenge in face recognition. As opposed to current techniques for pose invariant recognition, which either directly extract features or first normalize profile images frontal before feature extraction, we argue that it more desirable perform both tasks jointly allow them benefit from each other. To this end, propose a Invariant Model (PIM) recognition the wild, with three distinct novelties. First, PIM novel and unified deep architecture, containing Face...
Skeleton based action recognition distinguishes human actions using the trajectories of skeleton joints, which provide a very good representation for describing actions. Considering that recurrent neural networks (RNNs) with Long Short-Term Memory (LSTM) can learn feature representations and model long-term temporal dependencies automatically, we propose an end-to-end fully connected deep LSTM network recognition. Inspired by observation co-occurrences joints intrinsically characterize...
Human action analytics has attracted a lot of attention for decades in computer vision. It is important to extract discriminative spatio-temporal features model the spatial and temporal evolutions different actions. In this paper, we propose explore human recognition detection from skeleton data. We build our networks based on recurrent neural with long short-term memory units. The learned capable selectively focusing joints skeletons within each input frame paying levels outputs frames. To...