A5100781631- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Handwritten Text Recognition Techniques
- Advanced Vision and Imaging
- Video Analysis and Summarization
- Image Retrieval and Classification Techniques
- Robotics and Sensor-Based Localization
- Cryptography and Data Security
- Image Processing and 3D Reconstruction
- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Advanced Image Processing Techniques
- Visual Attention and Saliency Detection
- Adversarial Robustness in Machine Learning
- Gait Recognition and Analysis
- Multimodal Machine Learning Applications
- Anomaly Detection Techniques and Applications
- Vehicle License Plate Recognition
- Digital Filter Design and Implementation
- Geophysical and Geoelectrical Methods
- Advanced Image Fusion Techniques
- COVID-19 diagnosis using AI
- Computer Graphics and Visualization Techniques
- Advanced Computational Techniques and Applications
Ningbo Medical Center Lihuili Hospital
2025
Peking University
2015-2024
Chengdu University of Technology
2024
Xidian University
2024
China National Petroleum Corporation (China)
2008-2023
Xinjiang University
2022-2023
Dalian University of Technology
2023
Shandong Institute of Automation
2019-2022
Convergence
2022
Huazhong University of Science and Technology
2007-2020
The robust detection of small targets is one the key techniques in infrared search and tracking applications. A novel target method a single image proposed this paper. Initially, traditional model generalized to new patch-image using local patch construction. Then, because non-local self-correlation property background image, based on formulated as an optimization problem recovering low-rank sparse matrices, which effectively solved stable principle component pursuit. Finally, simple...
Feature pyramids are widely exploited by both the state-of-the-art one-stage object detectors (e.g., DSSD, RetinaNet, RefineDet) and two-stage Mask RCNN, DetNet) to alleviate problem arising from scale variation across instances. Although these with feature achieve encouraging results, they have some limitations due that only simply construct pyramid according inherent multiscale, pyramidal architecture of backbones which originally designed for classification task. Newly, in this work, we...
In existing CNN based detectors, the backbone network is a very important component for basic feature1 extraction, and performance of detectors highly depends on it. this paper, we aim to achieve better detection by building more powerful from ones like ResNet ResNeXt. Specifically, propose novel strategy assembling multiple identical backbones composite connections between adjacent backbones, form named Composite Backbone Network (CBNet). way, CBNet iteratively feeds output features...
Fusing the camera and LiDAR information has become a de-facto standard for 3D object detection tasks. Current methods rely on point clouds from sensor as queries to leverage feature image space. However, people discovered that this underlying assumption makes current fusion framework infeasible produce any prediction when there is malfunction, regardless of minor or major. This fundamentally limits deployment capability realistic autonomous driving scenarios. In contrast, we propose...
Modern top-performing object detectors depend heavily on backbone networks, whose advances bring consistent performance gains through exploring more effective network structures. In this paper, we propose a novel and flexible framework, namely CBNet, to construct high-performance using existing open-source pre-trained backbones under the pre-training fine-tuning paradigm. particular, CBNet architecture groups multiple identical backbones, which are connected composite connections....
We propose an end-to-end saliency-guided deep neural network (SGDNet) for no-reference image quality assessment (NR-IQA). Our SGDNet is built on multi-task learning framework in which two sub-tasks including visual saliency prediction and are jointly optimized with a shared feature extractor. The existing CNN-based NR-IQA methods usually consider distortion identification as the auxiliary sub-task cannot accurately identify complex mixtures of distortions exist authentically distorted...
Malicious applications of deepfakes (i.e., technologies generating target facial attributes or entire faces from images) have posed a huge threat to individuals' reputation and security. To mitigate these threats, recent studies proposed adversarial watermarks combat deepfake models, leading them generate distorted outputs. Despite achieving impressive results, low image-level model-level transferability, meaning that they can protect only one image specific model. address issues, we propose...
To achieve autonomous driving, developing 3D detection fusion methods, which aim to fuse the camera and LiDAR information, has draw great research interest in recent years. As a common practice, people rely on large-scale datasets fairly compare performance of different methods. While these have been carefully cleaned ideally minimize any potential noise, we observe that they cannot truly reflect data seen real vehicle, whose tends be noisy due various reasons. This hinders ability simply...
Recently, neural architecture search (NAS) has been exploited to design feature pyramid networks (FPNs) and achieved promising results for visual object detection. Encouraged by the success, we propose a novel One-Shot Path Aggregation Network Architecture Search (OPANAS) algorithm, which significantly improves both searching efficiency detection accuracy. Specifically, first introduce six heterogeneous information paths build our space, namely top-down, bottom-up, fusing-splitting,...
Compared to query-based black-box attacks, transferbased attacks do not require any information of the attacked models, which ensures their secrecy. However, most existing transfer-based approaches rely on ensembling multiple models boost attack transferability, is time- and resource-intensive, mention difficulty obtaining diverse same task. To address this limitation, in work, we focus single-model object detection, utilizing only one model achieve a high-transferability adversarial...
Dynamic neural network is an emerging research topic in deep learning. With adaptive inference, dynamic models can achieve remarkable accuracy and computational efficiency. However, it challenging to design a powerful detector, because of no suitable architecture exiting criterion for object detection. To tackle these difficulties, we propose framework detection, named DynamicDet. Firstly, carefully based on the nature detection task. Then, router analyze multi-scale information decide...
Face detection of comic characters is a necessary step in most applications, such as character retrieval, automatic classification and analysis. However, the existing methods were developed for simple cartoon images or small size datasets, performance remains to be improved. In this paper, we propose Faster R-CNN based method face characters. Our contribution twofold. First, binary task detection, empirically find that sigmoid classifier shows slightly better than softmax classifier. Second,...
Nowadays, general object detectors like YOLO and Faster R-CNN as well their variants are widely exploited in many applications. Many works have revealed that these extremely vulnerable to adversarial patch attacks. The perturbed regions generated by previous patch-based attack on very large which not necessary for attacking perceptible human eyes. To generate much less but more efficient perturbation, we propose a novel method detectors. Firstly, selection refining scheme find the pixels...