A5078550858- Advanced Vision and Imaging
- Computer Graphics and Visualization Techniques
- Advanced Image Processing Techniques
- 3D Shape Modeling and Analysis
- Image Enhancement Techniques
- Acoustic Wave Resonator Technologies
- Photorefractive and Nonlinear Optics
- Gait Recognition and Analysis
- Human Pose and Action Recognition
- Image and Signal Denoising Methods
- Photonic and Optical Devices
- Electronic Packaging and Soldering Technologies
- Image Processing and 3D Reconstruction
- Radio Frequency Integrated Circuit Design
- Advanced Power Amplifier Design
- Video Surveillance and Tracking Methods
- Vehicle License Plate Recognition
- Infrared Thermography in Medicine
- Photoacoustic and Ultrasonic Imaging
- Ultrasound Imaging and Elastography
- Generative Adversarial Networks and Image Synthesis
- Hand Gesture Recognition Systems
- Human Motion and Animation
- Infrastructure Maintenance and Monitoring
- Elevator Systems and Control
University of Hong Kong
2021-2025
University of Electronic Science and Technology of China
2008-2020
China Special Equipment Inspection and Research Institute
2020
The Wallace H. Coulter Department of Biomedical Engineering
2018
Georgia Institute of Technology
2018
We present a new deep point cloud rendering pipeline through multi-plane projections. The input to the network is raw of scene and output are image or sequences from novel view along camera trajectory. Unlike previous approaches that directly project features 3D points onto 2D domain, we propose these into layered volume frustum. In this way, visibility can be automatically learnt by network, such ghosting effects due false check as well occlusions caused noise interferences both avoided...
Virtual environments (VEs) are pivotal for virtual, augmented, and mixed reality systems. Despite advances in 3D generation reconstruction, the direct creation of objects within an established scene (represented as NeRF) novel VE remains a relatively unexplored domain. This process is complex, requiring not only high-quality but also their seamless integration into existing scene. To this end, we propose pipeline featuring intuitive interface, dubbed GO-NeRF. Our approach takes text prompts...
Moiré patterns, appearing as color distortions, severely degrade image and video qualities when filming a screen with digital cameras. Considering the increasing demands for capturing videos, we study how to remove such undesirable moiré patterns in namely demoiréing. To this end, introduce first hand-held demoiréing dataset dedicated data collection pipeline ensure spatial temporal alignments of captured data. Further, baseline model implicit feature space alignment selective aggregation is...
This paper presents a new text-guided 3D shape generation approach DreamStone that uses images as stepping stone to bridge the gap between text and modalities for generating shapes without requiring paired data. The core of our is two-stage feature-space alignment strategy leverages pre-trained single-view reconstruction (SVR) model map CLIP features shapes: begin with, image feature detail-rich space SVR model, then through encouraging CLIP-consistency rendered input text. Besides, extend...
We propose a Generative Adversarial Network (GAN)-based architecture for achieving high-quality physically based rendering (PBR). Conventional PBR relies heavily on ray tracing, which is computationally expensive in complicated environments. Some recent deep learning-based methods can improve efficiency but cannot deal with illumination variation well. In this paper, we PBR-GAN, an end-to-end GAN-based network that solves these problems while generating natural photo-realistic images. Two...
Railway patrolling inspection train has been widely used for railway infrastructure safety monitoring. Cameras are mounted on the train, which can capture image of overhead contact power line system defect detection. In catenary support device system, insulator keep equipment insulated from other equipment. Defect detection insulators is extremely important to safety. recent years, some achievements have made in based computer vision. We propose an localization algorithm and using deep...
Physically based rendering has been widely used to generate photo-realistic images, which greatly impacts industry by providing appealing rendering, such as for entertainment and augmented reality, academia serving large scale high-fidelity synthetic training data hungry methods like deep learning. However, physically heavily relies on ray-tracing, can be computational expensive in complicated environment hard parallelize. In this paper, we propose an end-to-end learning approach...
Abstract:
It is especially challenging to achieve real-time human motion tracking on a standalone VR Head-Mounted Display (HMD) such as Meta Quest and PICO. In this paper, we propose HMD-Poser, the first unified approach recover full-body motions using scalable sparse observations from HMD body-worn IMUs. particular, it can support variety of input scenarios, HMD, HMD+2IMUs, HMD+3IMUs, etc. The scalability inputs may accommodate users' choices for both high accuracy easy-to-wear. A lightweight...
Scene reconstruction from multi-view images is a fundamental problem in computer vision and graphics. Recent neural implicit surface methods have achieved high-quality results; however, editing manipulating the 3D geometry of reconstructed scenes remains challenging due to absence naturally decomposed object entities complex object/background compositions. In this paper, we present Total-Decom, novel method for with minimal human interaction. Our approach seamlessly integrates Segment...
This paper presented an improved design on a 121.4 MHz overtone TCXO which exhibited low phase noise. The temperature compensation approach differs from conventional ones such as adding series inductances or frequency multiplication. It makes of 100 5th crystal oscillator mixed with the 21.4 fundamental mode voltage controlled (VCXO). And then, was filtered and amplified to produce output. For better performance, computer simulation tool Agilent ADS used estimate noise while designing these...
This paper presented a new 121.4 MHz overtone TCXO with low phase noise. The compensation approach differs from conventional ones such as adding series inductances or frequency multiplication. It can provide very high stability and noise level for an oscillator. makes the of 100 5 th crystal oscillator mixed that 21.4 fundamental mode voltage controlled (VCXO). And then, was filtered amplified to produce output. A microcontroller used control compensating VCXO at given temperature generate...
We present a new deep point cloud rendering pipeline through multi-plane projections. The input to the network is raw of scene and output are image or sequences from novel view along camera trajectory. Unlike previous approaches that directly project features 3D points onto 2D domain, we propose these into layered volume frustum. In this way, visibility can be automatically learnt by network, such ghosting effects due false check as well occlusions caused noise interferences both avoided...
We introduce a novel video denoising approach which can produce clean by utilizing redundant image patches existed in the frames. Previous multi-frame denosing approaches either require registration or employ Patch Match algorithms for discovery of patch redundancy. However, these computations are time-consuming and prone to errors. On other hand, nearly all captured videos have been compressed. Such compression rich set block-based motion vectors that be utilized extraction, leading...
Rendering plays an important role in many fields such as virtual reality and film, but the high dependence on computing sources human experience hinders its application. With development of deep learning, neural rendering has attracted much attention due to impressive performance efficiency than traditional rendering. In this paper, we mainly introduce two works, one is simulation other image-based novel view Moreover, also discuss potential applications (i.e. data augmentation) based...