A5100396713- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Medical Image Segmentation Techniques
- Image Enhancement Techniques
- Optical Coatings and Gratings
- Image and Signal Denoising Methods
- Advanced Measurement and Detection Methods
- Machine Learning and ELM
- Advanced MRI Techniques and Applications
- 3D Shape Modeling and Analysis
- Image Processing Techniques and Applications
- Computer Graphics and Visualization Techniques
- Image Retrieval and Classification Techniques
- Radio Wave Propagation Studies
- Speech and Audio Processing
- Generative Adversarial Networks and Image Synthesis
- Icing and De-icing Technologies
- Face and Expression Recognition
- Thermal Radiation and Cooling Technologies
University of Electronic Science and Technology of China
2014-2024
Southeast University
2024
Shanghai University of Electric Power
2024
Ocean University of China
2024
National Engineering Research Center of Electromagnetic Radiation Control Materials
2014
Few-shot segmentation (FSS) aims to segment the novel class with a few annotated images. Due CLIP's advantages of aligning visual and textual information, integration CLIP can enhance generalization ability FSS model. However, even model, existing CLIP-based methods are still subject biased prediction towards base class, which is caused by class-specific feature level interactions. To solve this issue, we propose Prior Guided Mask Assemble Network (PGMA-Net). It employs class-agnostic mask...
The Diffusion model has a strong ability to generate wild images. However, the can just inaccurate images with guidance of text, which makes it very challenging directly apply text-guided generative for virtual try-on scenarios. Taking as guiding conditions diffusion model, this paper proposes brand new personalized (PE-VITON), uses two stages (shape control and texture guidance) decouple clothing attributes. Specifically, proposed adaptively matches human body parts through Shape Control...
Abstract Background Medical imaging plays a pivotal role in the real‐time monitoring of patients during diagnostic and therapeutic processes. However, clinical scenarios, acquisition multi‐modal protocols is often impeded by number factors, including time economic costs, cooperation willingness patients, quality, even safety concerns. Purpose We proposed learning‐based medical image synthesis method to simplify multi‐contrast MRI. Methods redesigned basic structure Mamba block explored...
Class activation map (CAM) highlights regions of classes based on classification network, which is widely used in weakly supervised tasks. However, it faces the problem that class are usually small and local. Although several efforts paid to second step (the CAM generation step) have partially enhanced generation, we believe such also caused by first (training step), because single model trained entire contains finite discriminate information limits object region extraction. To this end,...
Compared with electromagnetic compatibility (EMC) testing in anechoic rooms, open-area EMC takes advantage of situ and engine running status measurement but suffers from non-negligible external interference. This paper proposes a novel environmental interference suppression method (named the algorithm (E2ISA)) that separates signals backgrounds via image segmentation recognizes near–far site signal group time-varying features based on difference near-site EM radiative characteristic. We find...
Object detection in unmanned aerial vehicle (UAV) images has recently become a popular tool with widespread applications, but due to the high altitude of UAV and subtle inter-class differences, effectively detecting object locations achieving accurate classification remain challenge. Existing methods often rely on multi-scale feature fusion attention mechanisms address these problems. However, tend significantly increase computational overhead introduce unwanted background noise interference...
Few-shot segmentation segments object regions of new classes with a few manual annotations. Its key step is to establish the transformation module between support images (annotated images) and query (unlabeled images), so that cues can guide images. The existing methods form model based on global cues, which however ignores local are verified in this paper be very important for transformation. This proposes where relationship features used To enhance generalization performance network,...