A5101470618- Advanced Vision and Imaging
- Visual Attention and Saliency Detection
- Advanced Image Processing Techniques
- Ocular Surface and Contact Lens
- Video Surveillance and Tracking Methods
- Advanced Optical Sensing Technologies
- Optical measurement and interference techniques
- Image Enhancement Techniques
- Advanced Image Fusion Techniques
- Computer Graphics and Visualization Techniques
- Fault Detection and Control Systems
- Advanced Steganography and Watermarking Techniques
- Advanced Statistical Methods and Models
- Image and Signal Denoising Methods
- Robotics and Sensor-Based Localization
- Ocular Diseases and Behçet’s Syndrome
- Law in Society and Culture
- Spectroscopy and Chemometric Analyses
- Digital Media Forensic Detection
- 3D Shape Modeling and Analysis
- Infrared Target Detection Methodologies
Hong Kong University of Science and Technology
2023
University of Hong Kong
2020-2023
Beijing Jiaotong University
2021
Tongji University
2015
Wuhan University
2015
We present an approach to predict future video frames given a sequence of continuous in the past. Instead synthesizing images directly, our is designed understand complex scene dynamics by decoupling background and moving objects. The appearance components predicted non-rigid deformation affine transformation anticipated appearances are combined create reasonable future. With this procedure, method exhibits much less tearing or distortion artifact compared other approaches. Experimental...
Due to the high similarity between camouflaged instances and background, recently proposed instance segmentation (CIS) faces challenges in accurate localization segmentation. To this end, inspired by query-based transformers, we propose a unified multi-task learning framework for segmentation, termed UQFormer, which builds set of mask queries boundary learn shared composed query representation efficiently integrates global object region cues, simultaneous detection scenarios. Specifically,...
Camouflaged object detection is a challenging task that aims to identify objects are highly similar their background. Due the powerful noise-to-image denoising capability of diffusion models, in this paper, we propose diffusion-based framework for camouflaged detection, termed diffCOD, new considers segmentation as process from noisy masks masks. Specifically, mask diffuses ground-truth random distribution, and designed model learns reverse noising process. To strengthen learning, input...
We present a novel approach to joint depth and normal estimation for time-of-flight (ToF) sensors. Our model learns predict the high-quality maps jointly from ToF raw sensor data. To achieve this, we meticulously constructed first large-scale dataset (named ToF-100) with paired data ground-truth high-resolution provided by an industrial camera. In addition, also design simple but effective framework estimation, applying robust Chamfer loss via jittering improve performance of our model....
We present a novel approach to joint depth and normal estimation for time-of-flight (ToF) sensors. Our model learns predict the high-quality maps jointly from ToF raw sensor data. To achieve this, we meticulously constructed first large-scale dataset (named ToF-100) with paired data ground-truth high-resolution provided by an industrial camera. In addition, also design simple but effective framework estimation, applying robust Chamfer loss via jittering improve performance of our model....
In robotic applications, we often obtain tons of 3D point cloud data without color information, and it is difficult to visualize clouds in a meaningful colorful way. Can colorize for better visualization? Existing deep learning-based colorization methods usually only take simple objects as input, their performance complex scenes with multiple limited. To this end, paper proposes novel semantics-and-geometry-aware network, termed SGNet, vivid scene-level colorization. Specifically, propose...
Due to the high similarity between camouflaged instances and background, recently proposed instance segmentation (CIS) faces challenges in accurate localization segmentation. To this end, inspired by query-based transformers, we propose a unified multi-task learning framework for segmentation, termed UQFormer, which builds set of mask queries boundary learn shared composed query representation efficiently integrates global object region cues, simultaneous detection scenarios. Specifically,...
We present an approach to predict future video frames given a sequence of continuous in the past. Instead synthesizing images directly, our is designed understand complex scene dynamics by decoupling background and moving objects. The appearance components predicted non-rigid deformation affine transformation anticipated appearances are combined create reasonable future. With this procedure, method exhibits much less tearing or distortion artifact compared other approaches. Experimental...