- Multimodal Machine Learning Applications
- Advanced Image and Video Retrieval Techniques
- Topic Modeling
- Human Pose and Action Recognition
- Chaos-based Image/Signal Encryption
- Cryptography and Data Security
- Coding theory and cryptography
- Advanced Neural Network Applications
- Face and Expression Recognition
- Cryptographic Implementations and Security
- Face recognition and analysis
- Inertial Sensor and Navigation
- Target Tracking and Data Fusion in Sensor Networks
- Explainable Artificial Intelligence (XAI)
- Video Analysis and Summarization
- Complexity and Algorithms in Graphs
- Natural Language Processing Techniques
- Anomaly Detection Techniques and Applications
- Cryptography and Residue Arithmetic
- Artificial Intelligence in Games
- Image and Signal Denoising Methods
- Cellular Automata and Applications
- Advanced Image Processing Techniques
- Advanced Computing and Algorithms
- Advanced Clustering Algorithms Research
Shandong Normal University
2025
Hubei Engineering University
2014-2024
Nanjing University of Aeronautics and Astronautics
2019-2024
Tencent (China)
2024
Chongqing University
2023
Huazhong University of Science and Technology
2021-2022
Shenzhen University
2021
Meizu (China)
2021
Luohe Medical College
2021
University of Utah
2018
Human-Object Interaction (HOI) detection is a fundamental task in high-level human-centric scene understanding. We propose PhraseHOI, containing HOI branch and novel phrase branch, to leverage language prior improve relation expression. Specifically, the supervised by semantic embeddings, whose ground truths are automatically converted from original annotations without extra human efforts. Meanwhile, label composition method proposed deal with long-tailed problem HOI, which composites labels...
Human-Object Interactions (HOI) detection, which aims to localize a human and relevant object while recognizing their interaction, is crucial for understanding still image. Recently, tranformer-based models have significantly advanced the progress of HOI detection. However, capability these has not been fully explored since Object Query model always simply initialized as just zeros, would affect performance. In this paper, we try study issue promoting transformer-based detectors by...
Accurate cancer survival prediction remains a critical challenge in clinical oncology, largely due to the complex and multi-omics nature of data. Existing methods often struggle capture comprehensive range informative features required for precise predictions. Here, we introduce PCLSurv, an innovative deep learning framework designed using PCLSurv integrates autoencoders extract omics-specific employs sample-level contrastive identify distinct yet complementary characteristics across data...
Neural networks models have gained unprecedented popularity in natural language processing due to their state-of-the-art performance and the flexible end-to-end training scheme. Despite advantages, lack of interpretability hinders deployment refinement models. In this work, we present a visualization library for creating customized visual analytic environments, which user can investigate interrogate relationships among input, model internals (i.e., attention), output predictions, turn shed...
During civil aviation flights, the aircraft needs to accurately monitor real-time navigation capability and determine whether onboard system performance meets required (RNP). The airborne flight management (FMS) uses actual (ANP) quantitatively calculate uncertainty of position estimation, its evaluation accuracy is highly dependent on estimation covariance matrix (PECM) provided by integrated system. This paper proposed an adaptive PECM method based variational Bayes (VB) solve problem ANP...
Non-blind image deblurring algorithms based on the regularization can cause ringing artifacts. Using hyper-Laplacian priors, Krishnan D has proposed a non-blind model which reduce artifacts well, but it damages edge details. To solve this problem, parameter adaptive regional division is in paper. It effectively overcome and maintain
Hash functions are the most important component for communication security protocols. Based on coupled map lattice together with traditional Merkle-Damgard iterated structure, a new hash function OCMLHash construction was proposed. Through defining float point number storage presentation and corresponding basic operations, proposed can complete chaos operation output value 32-bit word using operations such as shifter, adder etc. Compared existing chaotic functions, this cuts down time...
Shot boundary detection (SBD) plays an important role in video understanding, since most recent works take the shot as minimal granularity instead of frames for upstream tasks. However, large variations hard-cut and gradual-change transitions within shots significantly limit performance SBD. To deal with variations, we propose a multi-task architecture called Transnet++. Transnet++ disentangles two types transition adopts separate branches to predict them respectively. Two share same...
Human-Object Interaction (HOI) detection is a fundamental task in high-level human-centric scene understanding. We propose PhraseHOI, containing HOI branch and novel phrase branch, to leverage language prior improve relation expression. Specifically, the supervised by semantic embeddings, whose ground truths are automatically converted from original annotations without extra human efforts. Meanwhile, label composition method proposed deal with long-tailed problem HOI, which composites labels...
Dynamic Scene Graph Generation (DSGG) focuses on identifying visual relationships within the spatial-temporal domain of videos. Conventional approaches often employ multi-stage pipelines, which typically consist object detection, temporal association, and multi-relation classification. However, these methods exhibit inherent limitations due to separation multiple stages, independent optimization sub-problems may yield sub-optimal solutions. To remedy limitations, we propose a one-stage...
Abstract The opacity of deep convolutional neural network(CNN) models has hindered their performance enhancement across various domains, posing challenges in understanding internal mechanisms. To address this, computer vision developed approaches to assess CNN interpretability via visualization. However, existing techniques often encounter noise during gradient calculation and may produce rough, blurry saliency maps, leading the localization meaningless information. This paper proposes...
The integrity of airborne inertial navigation systems (INSs) is the key to ensuring safe flight civil aircraft. attitude and heading reference system (AHRS) introduced into construction a redundant system. As backup for an INS, AHRS exhibits different device performance. A sequential weighted generalized likelihood ratio test (SWGLT) method, based on principal component parity vector (PPV), proposed. PPV method improves adaptability detection threshold sensors’ noise probability correct...
Single Sample Per Person (SSPP) problem, which means that there is only one training sample for each gallery subject, a great challenge face recognition to date. In this work, we address problem by presenting novel framework combines specific learning and generic learning. The proposed approach directly inspired from the complementarity between former takes full advantage of samples attempts seek low-dimensional subspace can maximize class separability, while latter able provide...
NaSHA is a family of hash functions submitted by Markovski and Mileva, it accepted as one the first SHA-3 round candidates. In this paper, we present collision attack on for output sizes 384-bit 512-bit. This based weakness in generate course state words, fact that quasigroup operation used compression function are determined partial words. The time complexity about 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">128</sup> with negligible...
In this paper, a drag model-aided fault-tolerant state estimation method is presented for quadrotors. Firstly, the model accuracy was improved by modeling an angular rate related item and acceleration item, which are with flight maneuver. Then model, light detection ranging (LIDAR), inertial measurement unit (IMU) were fused based on Federal Kalman filter frame. filter, LIDAR fault detected isolated, disturbance to estimated compensated. Some experiments carried out, showing that velocity...
Certificate-based cryptography proposed by Gentry in Eurocrypt 2003 combines the advantages of traditional public key (PKI) and identity-based cryptography, removes certificate management problem private escrow security concern. Based on computational Diffie-Hellman assumption, a certificate-based signature scheme is constructed to insure communication mobile Ad hoc networks,. The proved under Random Oracle Model. also efficient, since signing algorithm does not need computation bilinear...
In order to study machine translation more in-depth, it is particularly important for the research of artificial intelligence with fuzzy algorithms convert an unfamiliar language into a mature language. The neural network model has been developed in recent years and achieved rich results. Aiming at current lack accuracy (NMT), which may cause ambiguity, this paper takes English as example proposes optimization based on theory. On basis NMT translation, first semantics classified, semantic...