A5109012644- Reinforcement Learning in Robotics
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- COVID-19 diagnosis using AI
- Advanced Neural Network Applications
- Artificial Intelligence in Games
- Video Surveillance and Tracking Methods
- Visual Attention and Saliency Detection
- Anomaly Detection Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Nasolacrimal Duct Obstruction Treatments
- Adversarial Robustness in Machine Learning
- Generative Adversarial Networks and Image Synthesis
- Evolutionary Algorithms and Applications
- Advanced Vision and Imaging
- Head and Neck Surgical Oncology
- Image and Video Quality Assessment
- Natural Language Processing Techniques
- Viral Infectious Diseases and Gene Expression in Insects
- Topic Modeling
- Data Stream Mining Techniques
- Digital Games and Media
- Human Pose and Action Recognition
- Image Enhancement Techniques
- Olfactory and Sensory Function Studies
Institute of Automation
2012-2024
University of Chinese Academy of Sciences
2017-2024
Chinese Academy of Sciences
2013-2024
Henan Provincial Eye Hospital
2014-2024
Shandong Institute of Automation
2014-2024
Beijing Academy of Artificial Intelligence
2019-2023
Xiangshan County First People's Hospital
2023
Ningbo First Hospital
2022
Yanshan University
2022
Wannan Medical College
2022
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...
Human can well recognize images of novel categories just after browsing few examples these categories. One possible reason is that they have some external discriminative visual information about from their prior knowledge. Inspired this, we propose a Knowledge Transfer Network architecture (KTN) for few-shot image recognition. The proposed KTN model jointly incorporates feature learning, knowledge inferring and classifier learning into one unified framework optimal compatibility. First, the...
It is common but challenging to address high-resolution image blending in the automatic photo editing application. In this paper, we would like focus on solving problem of blending, where composite images are provided. We propose a framework called Gaussian-Poisson Generative Adversarial Network (GP-GAN) leverage strengths classical gradient-based approach and Networks. To best our knowledge, it's first work that explores capability GANs task. Concretely, Equation formulate problem, which...
Zero-shot learning (ZSL) aims to recognize unseen image categories by an embedding space between and semantic representations. For years, among existing works, it has been the center task learn proper mapping matrices aligning visual space, whilst importance discriminative representations for ZSL is ignored. In this work, we retrospect methods demonstrate necessity both instances of ZSL. We propose end-to-end network that capable 1) automatically discovering regions a zoom network; 2) in...
Image cropping aims at improving the aesthetic quality of images by adjusting their composition. Most weakly supervised methods (without bounding box supervision) rely on sliding window mechanism. The mechanism requires fixed aspect ratios and limits region with arbitrary size. Moreover, method usually produces tens thousands windows input image which is very time-consuming. Motivated these challenges, we firstly formulate as a sequential decision-making process propose Aesthetics Aware...
It is common but challenging to address high-resolution image blending in the automatic photo editing application. In this paper, we would like focus on solving problem of blending, where composite images are provided. We propose a framework called Gaussian-Poisson Generative Adversarial Network (GP-GAN) leverage strengths classical gradient-based approach and Networks. To best our knowledge, it's first work that explores capability GANs task. Concretely, Equation formulate problem, which...
Building agents with adaptive behavior in cooperative tasks stands as a paramount goal the realm of multi-agent systems. Current approaches to developing rely primarily on learning-based methods, whose policy generalization depends heavily diversity teammates they interact during training phase. Such reliance, however, constrains agents' capacity for strategic adaptation when cooperating unfamiliar teammates, which becomes significant challenge zero-shot coordination scenarios. To address...
Localizing objects in cluttered backgrounds is challenging under large-scale weakly supervised conditions. Due to the image condition, usually have large ambiguity with backgrounds. Besides, there also a lack of effective algorithm for localization However, contain useful latent information, e.g., sky aeroplane class. If this information can be learned, object-background largely reduced and background suppressed effectively. In paper, we propose category learning (LCL) LCL an unsupervised...
Large Language Models (LLMs) have shown impressive reasoning capabilities in well-defined problems with clear solutions, such as mathematics and coding. However, they still struggle complex real-world scenarios like business negotiations, which require strategic reasoning-an ability to navigate dynamic environments align long-term goals amidst uncertainty. Existing methods for face challenges adaptability, scalability, transferring strategies new contexts. To address these issues, we propose...
To recognize objects of the unseen classes, most existing Zero-Shot Learning(ZSL) methods first learn a compatible projection function between common semantic space and visual based on data source seen then directly apply it to target classes. However, in real scenarios, distribution domain might not match well, thus causing well-known \textbf{domain shift} problem. Based observation that features test instances can be separated into different clusters, we propose new structure constraint...
Image cropping aims at improving the quality of images by removing unwanted outer areas, which is widely used in photography and printing industry. Most previous methods that don't need bounding box supervision rely on sliding window mechanism. The method results fixed aspect ratios limits shape region. Moreover, usually produces lots candidates input image, very time-consuming. Motivated these challenges, we formulate image as a sequential decision-making process propose reinforcement...
Despite the impressive performance of deep learning models, they suffer from catastrophic forgetting, which refers to a significant decline in overall when trained with new classes added incrementally. The primary reason for this phenomenon is overlapping or confusion between feature space representations old and classes. In study, we examine issue propose model that can mitigate problem by more transferable features. We employ contrastive learning, recent breakthrough learn visual better...
Indoor RGB-D semantic segmentation is a new and challenging problem. Traditional methods usually apply two-stream convolutional neural networks (CNNs) to represent RGB depth images respectively, fuse the two streams on specific layer. In this paper, we explore several fusion strategies based two-stream-CNN framework point out such single-layer method cannot exploit complementary cues well for segmentation. To address problem, propose novel Semantics-guided Multi-level feature approach, which...
Finding views with a good composition from an input image is common but challenging problem. There are usually at least dozens of candidates (regions) in image, and how to evaluate these subjective. Most existing methods only use the feature corresponding each candidate quality. However, mutual relations between play essential role composing shot due comparative nature this Motivated by this, we propose graph-based module gated update model different candidates. The region features...
Online Continual Learning (OCL), as a core step towards achieving human-level intelligence, aims to incrementally learn and accumulate novel concepts from streaming data that can be seen only once, while alleviating catastrophic forgetting on previously acquired knowledge. Under this mode, the model needs new classes or tasks in an online manner, distribution may change over time. Moreover, task boundaries identities are not available during training evaluation. To balance stability...
Background: Aberrant expression of circular RNA (circRNA) is involved in the occurrence and development multifarious cancers, including oral squamous cell carcinoma (OSCC). However, biological role circGDI2 action mechanism OSCC remain largely unclear. Methods: The levels circGDI2, miR-454-3p forkhead box F2 (FOXF2) were examined by quantitative real-time PCR (qRT-PCR) or Western blot. stability was confirmed Ribonuclease R (RNase R) assay. Cell Counting Kit 8 (CCK8) assay, colony formation...