A5100376572- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Silicon and Solar Cell Technologies
- Anomaly Detection Techniques and Applications
- Semiconductor materials and interfaces
- Thin-Film Transistor Technologies
- Advanced Vision and Imaging
- Supercapacitor Materials and Fabrication
- Visual Attention and Saliency Detection
- Advanced Image Processing Techniques
- Advanced battery technologies research
- COVID-19 diagnosis using AI
- Video Surveillance and Tracking Methods
- Image and Video Quality Assessment
- Radiopharmaceutical Chemistry and Applications
- 3D Shape Modeling and Analysis
- 3D Surveying and Cultural Heritage
- Nanowire Synthesis and Applications
- Video Coding and Compression Technologies
- Cancer Immunotherapy and Biomarkers
- Gait Recognition and Analysis
- Advancements in Battery Materials
Lanzhou University
2016-2025
University of Science and Technology of China
2016-2025
Peking University
2021-2025
Peking University Cancer Hospital
2021-2025
National Medical Products Administration
2022-2025
University of Electronic Science and Technology of China
2025
Sun Yat-sen University
2019-2024
Shanghai Children's Hospital
2021-2024
Shanghai Jiao Tong University
2024
Ministry of Education of the People's Republic of China
2024
Conventionally, deep neural networks are trained offline, relying on a large dataset prepared in advance. This paradigm is often challenged real-world applications, e.g. online services that involve continuous streams of incoming data. Recently, incremental learning receives increasing attention, and considered as promising solution to the practical challenges mentioned above. However, it has been observed subject fundamental difficulty -- catastrophic forgetting, namely adapting model new...
While feedforward deep convolutional neural networks (CNNs) have been a great success in computer vision, it is important to note that the human visual cortex generally contains more feedback than connections. In this paper, we will briefly introduce background of feedbacks cortex, which motivates us develop computational mechanism networks. addition inference traditional networks, loop introduced infer activation status hidden layer neurons according "goal" network, e.g., high-level...
Recent research on super-resolution has achieved great success due to the development of deep convolutional neural networks (DCNNs). However, arbitrary scale factor been ignored for a long time. Most previous researchers regard differentscale factors as independent tasks. They train specific model each which is inefficient in computing, and prior work only take several integer into consideration. In this work,we propose novel method called Meta-SR firstly solve (including non-integer...
3D petal-like NiCo<sub>2</sub>S<sub>4</sub> nanostructures have been fabricated <italic>via</italic> a simple, mild and efficient hydrothermal strategy the growth mechanism of nano-petals has investigated.
Abstract Wearable textile energy storage systems are rapidly growing, but obtaining carbon fiber fabric electrodes with both high capacitances to provide a density and mechanical strength allow the material be weaved or knitted into desired devices remains challenging. In this work, N/O‐enriched cloth large surface area pore volume is fabricated. An electrochemical oxidation method used modify chemistry through incorporation of active functional groups further increase specific cloth. The...
We propose a simple and application-friendly network (called SimpleNet) for detecting localizing anoma-lies. SimpleNet consists of four components: (1) pre-trained Feature Extractor that generates local features, (2) shallow Adapter transfers features to-wards target domain, (3) Anomaly Gener-ator counterfeits anomaly by adding Gaussian noise to normal (4) binary Discriminator distinguishes from features. During inference, the Generator would be discarded. Our approach is based on three...
The existing weakly supervised semantic segmentation (WSSS) methods pay much attention to generating accurate and complete class activation maps (CAMs) as pseudo-labels, while ignoring the importance of training networks. In this work, we observe that there is an inconsistency between quality pseudo-labels in CAMs performance final model, mislabeled pixels mainly lie on boundary areas. Inspired by these findings, argue focus WSSS should be shifted robust learning given noisy further propose...
In this paper, we propose a novel domain-specific dataset named VegFru for fine-grained visual categorization (FGVC). While the existing datasets FGVC are mainly focused on animal breeds or man-made objects with limited labelled data, is larger consisting of vegetables and fruits which closely associated daily life everyone. Aiming at domestic cooking food management, categorizes according to their eating characteristics, each image contains least one edible part same usage. Particularly,...
Temporal action detection is a challenging task due to vagueness of boundaries. To tackle this issue, we propose an end-to-end progressive boundary refinement network (PBRNet) in paper. PBRNet belongs the family one-stage detectors and equipped with three cascaded modules for localizing more precisely. Specifically, mainly consists coarse pyramidal detection, refined fine-grained detection. The first two build feature pyramids perform anchor-based third one explores frame-level features...
Recent years have witnessed great progress of object detection. However, due to the domain shift problem, applying knowledge an detector learned from one specific another often suffers severe performance degradation. Most existing methods adopt feature alignment either on backbone network or instance classifier increase transferability detector. Differently, we propose perform in RPN stage such that foreground and background proposals target can be effectively distinguished. Specifically,...
A new composite electrode design was successfully fabricated based on 3D flexible graphene foams (GF) with interconnected macropores as the freestanding substrate and a of MnO<sub>2</sub> nanoparticles polypyrrole (PPy) an integrated electrode.
In this work, we propose a novel method named Weighted Channel Dropout (WCD) for the regularization of deep Convolutional Neural Network (CNN). Different from which randomly selects neurons to set zero in fully-connected layers, WCD operates on channels stack convolutional layers. Specifically, consists two steps, i.e., Rating Channels and Selecting Channels, three modules, Global Average Pooling, Random Selection Number Generator. It filters according their activation status can be plugged...
Person re-identification aims at identifying a certain pedestrian across non-overlapping multi-camera networks in different time and places. Existing person approaches mainly focus on matching pedestrians images; however, little attention has been paid to re-identify videos. Compared images, video clips contain motion patterns of pedestrians, which is crucial re-identification. Moreover, consecutive frames present appearance with body poses from viewpoints, providing valuable information...
Claudin18.2 (CLDN18.2) is a tight junction protein that overexpressed in variety of solid tumors such as gastrointestinal cancer and oesophageal cancer. It has been identified promising target potential biomarker to diagnose tumor, evaluate efficacy, determine patient prognosis. TST001 recombinant humanized CLDN18.2 antibody selectively binds the extracellular loop human Claudin18.2. In this study, we constructed radionuclide zirconium-89 (89Zr) labled-TST001 detect expression stomach...
Many recent image restoration methods use Transformer as the backbone network and redesign blocks. Differently, we explore parameter-sharing mechanism over blocks propose a dynamic recursive process to address super-resolution task efficiently. We firstly present Recursive Image Super-resolution (RIST). By sharing weights across different blocks, plain forward through whole can be folded into iterations block. Such based not only reduce model size greatly, but also enable restoring images...
In this work we propose a novel framework named Dual-Net aiming at learning more accurate representation for image recognition. Here two parallel neural networks are coordinated to learn complementary features and thus wider network is constructed. Specifically, logically divide an end-to-end deep convolutional into functional parts, i.e., feature extractor classifier. The extractors of subnetworks placed side by side, which exactly form the DualNet. Then two-stream aggregated final...
Although nanostructured Ni/Co sulfides have shown compelling evidence and outstanding characteristics when applied for energy storage, their practical applications still face a challenge due to the sluggish electronic ionic transport at levels of mass loading. Herein, we report Ni–Co–S material with unique structure hollow nanospheres covered by interconnected nanosheets synthesized through an electrodeposition method. By systematically investigating effects Ni Co ions on materials growth...
Few-shot classification aims to categorize the samples from unseen classes with only few labeled samples. To address such a challenge, many methods exploit base set consisting of massive learn an instance embedding function, i.e., image feature extractor, and it is expected possess good transferability among different tasks. Such characteristics few-shot learning are essentially that traditional pursuing get discriminative representations. In this paper, we propose intact features by...
Combining electron- and hole-selective materials in one crystalline silicon (Si) solar cell, thereby avoiding any dopants, is not considered for application to photovoltaic industry until only comparable efficiency stable performance are achievable. Here, it demonstrated how a conventionally unstable electron-selective contact (ESC) optimized with huge boost stability as well improved electron transport. With the introduction of Ti thin film between a-Si:H(i)/LiF Al electrode, high-level...
This work focuses on a practical knowledge transfer task defined as Source-Free Unsupervised Domain Adaptation (SFUDA), where only well-trained source model and unlabeled target data are available. To fully utilize knowledge, we propose to the class relationship, which is domain-invariant but still under-explored in previous works. this end, first regard classifier weights of prototypes compute then novel probability-based similarity between target-domain samples by embedding source-domain...