A5101497048- Multimodal Machine Learning Applications
- Nuclear Engineering Thermal-Hydraulics
- Domain Adaptation and Few-Shot Learning
- Heat transfer and supercritical fluids
- COVID-19 diagnosis using AI
- Analog and Mixed-Signal Circuit Design
- Image Retrieval and Classification Techniques
- Nuclear reactor physics and engineering
- Natural Language Processing Techniques
- Human Pose and Action Recognition
- Atmospheric Ozone and Climate
- Digital Storytelling and Education
- Sulfur-Based Synthesis Techniques
- Topic Modeling
- 3D Modeling in Geospatial Applications
- Climate variability and models
- Advanced Memory and Neural Computing
- Radio Frequency Integrated Circuit Design
- Advancements in Semiconductor Devices and Circuit Design
- Safety and Risk Management
- Low-power high-performance VLSI design
- Text and Document Classification Technologies
- Ferroelectric and Negative Capacitance Devices
- Speech and dialogue systems
- Semiconductor materials and devices
Third Affiliated Hospital of Guangzhou Medical University
2024
Guangzhou Medical University
2024
Zhejiang University
2023-2024
Sun Yat-sen University
2022-2023
Fuzhou University
2023
Guangdong University of Technology
2023
Columbia University
2023
Beijing Information Science & Technology University
2023
Chengdu University of Information Technology
2021
China Meteorological Administration
2021
Weakly supervised semantic segmentation (WSSS) with image-level labels is a challenging task. Mainstream approaches follow multi-stage framework and suffer from high training costs. In this paper, we explore the potential of Contrastive Language-Image Pre-training models (CLIP) to localize different categories only without further training. To efficiently generate high-quality masks CLIP, propose novel WSSS called CLIP-ES. Our improves all three stages special designs for CLIP: 1) We...
Contrastive Language-Image Pre-training (CLIP) has demonstrated impressive capabilities in open-vocabulary classification. The class token the image encoder is trained to capture global features distinguish different text descriptions supervised by contrastive loss, making it highly effective for single-label However, shows poor performance on multi-label datasets because feature tends be dominated most prominent and nature of softmax operation aggravates it. In this study, we observe that...
In this paper, we explore a principal way to enhance the quality of widely pre-existing coarse masks, enabling them serve as reliable training data for segmentation models reduce annotation cost. contrast prior refinement techniques that are tailored specific or tasks in close-world manner, propose SAMRefiner, universal and efficient approach by adapting SAM mask task. The core technique our model is noise-tolerant prompting scheme. Specifically, introduce multi-prompt excavation strategy...
In recent years, the traditional simulation-based medical teaching approach has faced challenges in meeting requirements of practical emergency medicine education. This study utilized open-source tools and software to develop immersive panoramic videos using virtual reality technology for teaching. It aims investigate efficacy this novel methodology. transformation shifted focus from physical simulation education, establishing a metaverse
This paper presents ConvBench, a novel multi-turn conversation evaluation benchmark tailored for Large Vision-Language Models (LVLMs). Unlike existing benchmarks that assess individual capabilities in single-turn dialogues, ConvBench adopts three-level multimodal capability hierarchy, mimicking human cognitive processes by stacking up perception, reasoning, and creativity. Each level focuses on distinct capability, mirroring the progression from basic perception to logical reasoning...
Large Vision-Language Models (LVLMs) show significant strides in general-purpose multimodal applications such as visual dialogue and embodied navigation. However, existing evaluation benchmarks cover a limited number of tasks testing rudimentary capabilities, falling short tracking LVLM development. In this study, we present MMT-Bench, comprehensive benchmark designed to assess LVLMs across massive requiring expert knowledge deliberate recognition, localization, reasoning, planning....
Abstract Using time‐slice experiments performed with version 4 of the Whole Atmosphere Community Climate Model (WACCM4) and satellite observations, we investigate hemispheric differences in, seasonal dependence of, water vapor transport into stratosphere in response to Indo‐Pacific warm pool (IPWP) sea‐surface temperature (SST) variations. Specifically, amplitude lower stratospheric (SWV) a warmer (cooler) IPWP (i.e., Niño [Niña]) occurs mainly boreal winter when SST forcing is identical for...
There are many problems in the arc welding repair of aluminum alloy parts, such as excessive pores, coarse grains, and low hardness. Introducing external energy excitation is an effective solution. In this paper, 7075 parts repaired using - laser shock forging technique. The effects laser-wire spacing, pulse frequency, current speed, V-groove depth on surface morphology, section geometric characteristics, porosity layer studied. Additionally, hardness microstructure formed by compared....
Recently, generative domain adaptation has achieved remarkable progress, enabling us to adapt a pre-trained generator new target domain. However, existing methods simply the single and are limited modality, either text-driven or image-driven. Moreover, they cannot maintain well consistency with source domain, which impedes inheritance of diversity. In this paper, we propose UniHDA, \textbf{unified} \textbf{versatile} framework for hybrid multi-modal references from multiple domains. We use...
Recent text-to-image (T2I) models have had great success, and many benchmarks been proposed to evaluate their performance safety. However, they only consider explicit prompts while neglecting implicit (hint at a target without explicitly mentioning it). These may get rid of safety constraints pose potential threats the applications these models. This position paper highlights current state T2I toward prompts. We present benchmark named ImplicitBench conduct an investigation on impacts with...
Multimodal Large Language Models (MLLMs) have made significant strides in visual understanding and generation tasks. However, generating interleaved image-text content remains a challenge, which requires integrated multimodal abilities. While the progress unified models offers new solutions, existing benchmarks are insufficient for evaluating these methods due to data size diversity limitations. To bridge this gap, we introduce GATE OpenING (OpenING), comprehensive benchmark comprising 5,400...
This paper proposed an ultra-low power time-domain temperature sensor circuit for IoT applications. It uses 2- T structures to generate reference voltage and complementary-to-absolute-temperature (CTAT) voltage. The produces current by using mirror charge the capacitor. Then of capacitor is compared with CTAT a temperature-dependent pulse. pulse width then digitized represent temperature. designed 0.1S $\mu$m CMOS process. simulation results show that it measures from −10°C 60°C 1V supply...
Abstract Experimental investigation has been carried out in a closed loop rectangular natural circulation facility having single heated macro channel dimensioned and built based on concepts of similarity scale down circuit similar to the primary Nuclear reactor. The study aims observe critically analyze thermal hydraulic parameters (Temperature Variation, Heating Power rises) as it affects transfer heat by NC flow along its at predetermined inlet subcooled set sustained 40°C, 50°C, 60°C 80°C...
Contrastive Language-Image Pre-training (CLIP) has demonstrated impressive capabilities in open-vocabulary classification. The class token the image encoder is trained to capture global features distinguish different text descriptions supervised by contrastive loss, making it highly effective for single-label However, shows poor performance on multi-label datasets because feature tends be dominated most prominent and nature of softmax operation aggravates it. In this study, we observe that...
This paper proposes a CMOS image sensor that can achieves imaging and energy harvesting simultaneously without introducing additional P-N junctions in the pirel array. The proposed pixel utilizes vertical N+P-well/DNW/P-sub structures as photodiodes based on standard 180 nm mixed-signal process. N+P-well is used for imaging, while P-well/DNW DNW/P-sub are with shorting P-well P-sub together. Moreover, traditional 4 T has been improved by using pairs switches zero-threshold NMOS source...
There are many problems in arc welding repair of aluminum alloy parts, such as excessive pores, coarse grains, and low hardness. Introducing external energy excitation is an effective solution. In this paper, 7075 parts repaired by - laser shock forging. The effects laser-wire spacing, pulse frequency, current speed, V-groove depth on surface morphology, section geometric characteristics porosity layer studied, hardness microstructure formed forging compared. results show that a fine...
Abstract This work presents a phase detector (PD) having dead-zone free and static offset improvement performance. The proposed inherits the low power consumption advantage of conventional using two true-single-phase clocking (TSPC) DFFs. It also effectively reduces offset, even in presence inevitable charge pump current mismatch. And problem TSPC PD is overcome by falling edge delay inverter. implemented standard 180nm CMOS technology. dimension PD’s layout 11μm×16μm. Post-layout simulation...
Can a pre-trained generator be adapted to the hybrid of multiple target domains and generate images with integrated attributes them? In this work, we introduce new task -- Few-shot Hybrid Domain Adaptation (HDA). Given source several domains, HDA aims acquire an that preserves all without overriding domain's characteristics. Compared (DA), offers greater flexibility versatility adapt generators more composite expansive domains. Simultaneously, also presents challenges than DA as have access...
With the scaling of devices in past few decades, a lot problems arise include short tunnel effect and significant power consumption, which sub-threshold slope is considerate factor. However, Sub-threshold Slope MOSFET limited to 60 mV/decade at room temperature due Boltzmann Tyranny. TFETs (Tunnelling Field Effect Transistors) are most promising field-effect transistors candidates owing its potential overcome degradation. This article illustrates main principles explains how can degradation...
Remote sensing image segmentation plays an important role in realizing intelligent city construction.The current mainstream networks effectively improve the effect of remote images by deeply mining rich texture and semantic features images.But there are still some problems such as rough results small target region poor edge contour segmentation.To overcome these three challenges, we propose improved model, referred to MRU-Net, which adopts U-Net architecture its backbone.Firstly,...