A5100334018- Multimodal Machine Learning Applications
- Image Enhancement Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Image Processing Techniques
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Human Pose and Action Recognition
- Image and Signal Denoising Methods
- Biometric Identification and Security
- Face recognition and analysis
- Advanced Computational Techniques and Applications
- Face and Expression Recognition
- Advanced Vision and Imaging
- Retinal Imaging and Analysis
- Advanced Decision-Making Techniques
- Advanced Image Fusion Techniques
- Hand Gesture Recognition Systems
- EEG and Brain-Computer Interfaces
- Sparse and Compressive Sensing Techniques
- Video Surveillance and Tracking Methods
- Caching and Content Delivery
- Video Analysis and Summarization
- Medical Image Segmentation Techniques
- Recommender Systems and Techniques
- Imbalanced Data Classification Techniques
Tea Research Institute
2025
Chinese Academy of Agricultural Sciences
2025
University of Electronic Science and Technology of China
2021-2024
Huazhong University of Science and Technology
2024
The Affiliated Yongchuan Hospital of Chongqing Medical University
2024
Chongqing Medical University
2024
China University of Mining and Technology
2016-2024
Ocean University of China
2024
Beijing University of Technology
2024
Energy Foundation
2024
Low-light image enhancement (LLIE) investigates how to improve illumination and produce normal-light images. The majority of existing methods low-light images via a global uniform manner, without taking into account the semantic information different regions. Without priors, network may easily deviate from region's original color. To address this issue, we propose novel semantic-aware knowledge-guided framework (SKF) that can assist model in learning rich diverse priors encapsulated...
Quercetin is a potent chemotherapeutic drug. Clinical trials exploring different schedules of administration quercetin have been hampered by its extreme water insolubility. To overcome this limitation, study aimed to develop liposomal and investigate distribution in vivo antitumor efficacy vitro.Quercetin was encapsulated polyethylene glycol 4000 liposomes. Biodistribution i.v. at 50 mg/kg tumor-bearing mice detected high-performance liquid chromatography. Induction apoptosis vitro tested....
In this article we propose a dynamical model with seven compartments to describe the transmission of COVID-19 in China. The home quarantine strategy has played vital role controlling disease spread. Based on Least-Squares procedure and officially published data, estimation parameters for proposed is obtained. control reproduction number most provinces China are analyzed. Attention that period must be long enough. Once removed, still high risk human-to-human continuously. study, comprehensive...
In this paper, we study the composed query image retrieval, which aims at retrieving target similar to query, i.e., a reference and desired modification text. Compared with conventional task is more challenging as it not only requires precisely aligning in common embedding space, but also simultaneously extracting related information from order properly extract existing methods usually embed vision-language inputs using different feature encoders, e.g., CNN for images LSTM/BERT text, then...
Despite the significant progress achieved in image de-raining by training an encoder-decoder network within image-to-image translation formulation, blurry results with missing details indicate deficiency of existing models. By interpreting as a conditional generator, which decoder acts generator conditioned on embedding learned encoder, unsatisfactory output can be attributed to low-quality encoder. In this paper, we hypothesize that there exists inherent mapping between latent optimal one,...
Deep metric learning has become a key component of cross-modal retrieval. By to pull the features matched instances closer while pushing mismatched farther away, one can learn highly robust multi-modal representations. Most existing retrieval methods leverage vanilla triplet loss train network, which cannot adaptively penalize pairs with different hardness. Although various weighting strategies have been designed for unimodal matching tasks, few applied tasks due specificity tasks. While are...
Video question answering~(Video-QA) is a task of answering natural language related to the content video. Existing methods generally explore single interactions between objects or frames, which are insufficient deal with sophisticated scenes in videos. To tackle this problem, we propose novel model, termed Progressive Graph Attention Network (PGAT), can jointly multiple visual relations on object-level, frame-level and clip-level. Specifically, object-level relation encoding, design two...
Abstract Background Chimeric antigen receptor (CAR) T cells and immune checkpoint blockades (ICBs) have made remarkable breakthroughs in cancer treatment, but the efficacy is still limited for solid tumors due to tumor heterogeneity microenvironment. The restrained treatment prompted us seek new potential therapeutic methods. Methods In this study, we conducted a small molecule compound library screen human BC cell line identify whether certain drugs contribute CAR killing. Signaling...
As an important visual understanding task, scene graph generation has been drawing widespread attention and could boost a broad range of downstream vision applications. Traditional methods based on different context refinements are trained with probabilistic chain rule, which treats objects relationships as independent entities. Despite their surprisingly great progress, such plain formulation unconsciously ignores the latent geometric structure entities relationships. To address this issue,...
Despite its significant progress, cross-modal retrieval still suffers from one-to-many matching cases, where the multiplicity of semantic instances in another modality could be acquired by a given query. However, existing approaches usually map heterogeneous data into learned space as deterministic point vectors. In spite their remarkable performance most similar instance, such embedding insufficient representation rich semantics correspondence. To address limitations, we intuitively extend...
Vein images generally appear darker with low contrast, which require contrast enhancement during preprocessing to design satisfactory hand vein recognition system. However, the modification introduced by (CE) is reported bring side effects through pixel intensity distribution adjustments. Furthermore, inevitable results of fake generation or information loss occur and make nearly all systems unconvinced. In this paper, a "CE-free" quality-specific system proposed, three improvements are...
Existing object detection models have been demonstrated to successfully discriminate and localize the predefined categories under seen or similar situations. However, open-world as required by autonomous driving perception systems refers recognizing unseen objects various scenarios. On one hand, knowledge gap between poses extreme challenges for trained with supervision only from categories. other domain differences across different scenarios also cause an additional urge take into...
Automatic detection and localization of objects in remote sensing images are great significance for systems. Existing frameworks usually train an object network using collected images. However, these models perform poorly due to the lack large-scale training datasets, which is often case special scenarios, e.g., ships open sea. Although image synthesis a common strategy alleviate issue data insufficiency, trained model still performs when being tested on real-world scenes. Aimed at this,...
Deep neural network (DNN) has demonstrated astounding performance in large-scale image recognition task, and pre-trained DNN models trained for one task have also been applied to domains different from their original purposes. Following such an idea, a novel hand-dorsa vein model is constructed by adopting on database as universal feature descriptor. Unlike most of these studies which adopt activations the fully connected layer representation, we convolutional region representation. However,...
Despite being highly secure, vein recognition suffers from the high inter-class similarity and intra-class variation resulting uncontrolled image capture, making design of discriminative robust representation very important. The recent success convolutional neural network (CNN) for various understanding tasks makes it a promising method feature extraction. However, limited variability in small-scale datasets leads to systems derived direct training or fine-tuning not transferable unreliable...