A5056443318- Retinal Imaging and Analysis
- Human Motion and Animation
- Human Pose and Action Recognition
- Advanced Vision and Imaging
- Retinopathy of Prematurity Studies
- 3D Shape Modeling and Analysis
- Computer Graphics and Visualization Techniques
- Retinal Diseases and Treatments
- Retinal and Optic Conditions
- Optical Coherence Tomography Applications
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Generative Adversarial Networks and Image Synthesis
- Neonatal Respiratory Health Research
- Neonatal and fetal brain pathology
- Multimodal Machine Learning Applications
- Cell Image Analysis Techniques
- Face and Expression Recognition
- Face recognition and analysis
- Image and Signal Denoising Methods
- Video Analysis and Summarization
- Robotics and Automated Systems
- Medical Image Segmentation Techniques
- Advanced Optical Imaging Technologies
- Neonatal Health and Biochemistry
Singapore National Eye Center
2025
Xuzhou Medical College
2025
Singapore Eye Research Institute
2025
Samsung (China)
2022-2024
Intel (United Kingdom)
2024
University of Utah
2021-2024
Adobe Systems (United States)
2021-2024
University of Illinois Urbana-Champaign
2023-2024
Hubei University of Chinese Medicine
2024
Hubei Provincial Hospital of Traditional Chinese Medicine
2024
In neural networks, it is often desirable to work with various representations of the same space. For example, 3D rotations can be represented quaternions or Euler angles. this paper, we advance a definition continuous representation, which helpful for training deep networks. We relate topological concepts such as homeomorphism and embedding. then investigate what are discontinuous 2D, 3D, n-dimensional rotations. demonstrate that rotations, all in real Euclidean spaces four fewer...
We present a real-time method for synthesizing highly complex human motions using novel training regime we call the auto-conditioned Recurrent Neural Network (acRNN). Recently, researchers have attempted to synthesize new motion by autoregressive techniques, but existing methods tend freeze or diverge after couple of seconds due an accumulation errors that are fed back into network. Furthermore, such only been shown be reliable relatively simple motions, as walking running. In contrast, our...
We present a novel method to realistically puppeteer and animate face from single RGB image using source video sequence. begin by fitting multilinear PCA model obtain the 3D geometry texture of target face. In order for animation be realistic, however, we need dynamic per-frame textures that capture subtle wrinkles deformations corresponding animated facial expressions. This problem is highly underconstrained, as cannot obtained directly image. Furthermore, if has closed mouth, it not...
A long-standing goal in computer vision is to capture, model, and realistically synthesize human behavior. Specifically, by learning from data, our enable virtual humans navigate within cluttered indoor scenes naturally interact with objects. Such embodied behavior has applications reality, games, robotics, while synthesized can be used as training data. The problem challenging because real motion diverse adapts the scene. For example, a person sit or lie on sofa many places varying styles....
With the ongoing pandemic, virtual concerts and live events using digitized performances of musicians are getting traction on massive multiplayer online worlds. However, well choreographed dance movements extremely complex to animate would involve an expensive tedious production process. In addition use motion capture systems, it typically requires a collaborative effort between animators, dancers, choreographers. We introduce complete system for synthesis, which can generate highly diverse...
Speckle noise is the main cause of poor optical coherence tomography (OCT) image quality. Convolutional neural networks (CNNs) have shown remarkable performances for speckle reduction. However, denoising still meets great challenges because deep learning-based methods need a large amount labeled data whose acquisition time-consuming or expensive. Besides, many CNNs-based design complex structure based with lots parameters to improve performance, which consume hardware resources severely and...
Drusen is considered as the landmark for diagnosis of AMD and important risk factor development AMD. Therefore, accurate segmentation drusen in retinal OCT images crucial early However, still very challenging due to large variations size shape drusen, blurred boundaries, speckle noise interference. Moreover, lack dataset with pixel-level annotation also a vital hindering improvement accuracy. To solve these problems, novel multi-scale transformer global attention network (MsTGANet) proposed...
Failure to recognize samples from the classes unseen during training is a major limitation of artificial intelligence in real-world implementation for recognition and classification retinal anomalies. We establish an uncertainty-inspired open set (UIOS) model, which trained with fundus images 9 conditions. Besides assessing probability each category, UIOS also calculates uncertainty score express its confidence. Our model thresholding strategy achieves F1 99.55%, 97.01% 91.91% internal...
In this paper we study the problem of shape analysis and its application in locating facial feature points on frontal faces. We propose a Bayesian inference solution based tangent approximation called model (BTSM). Similarity transform coefficients parameters BTSM are determined through MAP estimation. Tangent vector is treated as hidden state model, accordingly, an EM searching algorithm proposed to implement procedure. The major results our are: 1) updated by weighted average two vectors,...
Raw optical coherence tomography (OCT) images typically are of low quality because speckle noise blurs retinal structures, severely compromising visual and degrading performances subsequent image analysis tasks. In our previous study (Ma et al., 2018), we have developed a Conditional Generative Adversarial Network (cGAN) for removal in OCT collected by several commercial scanners, which collectively refer to as scanner T. this paper, improve the cGAN model apply it in-house (scanner B)...
Retinopathy of prematurity (ROP) is a retinal disease which frequently occurs in premature babies with low birth weight and considered as one the major preventable causes childhood blindness. Although automatic semi-automatic diagnoses ROP based on fundus image have been researched, most previous studies focused plus detection screening. There are few focusing staging, important for severity evaluation disease. To be consistent clinical 5-level novel effective deep neural network staging...
Choroidal neovascularization (CNV) is a typical symptom of age-related macular degeneration (AMD) and one the leading causes for blindness. Accurate segmentation CNV detection retinal layers are critical eye disease diagnosis monitoring. In this paper, we propose novel graph attention U-Net (GA-UNet) layer surface in optical coherence tomography (OCT) images. Due to deformation caused by CNV, it challenging existing models segment detect surfaces with correct topological order. We two...
The edge-detection problem is posed as one of detecting step discontinuities in the observed correlated image, using directional derivatives estimated with a random field model. Specifically, method consists representing pixels local window by 2-D causal autoregressive (AR) model, whose parameters are adaptively recursive least-squares algorithm. functions parameter estimates. An edge detected if second derivative direction maximum gradient negatively sloped and first estimate variance...
Learning latent representations of registered meshes is useful for many 3D tasks. Techniques have recently shifted to neural mesh autoencoders. Although they demonstrate higher precision than traditional methods, remain unable capture fine-grained deformations. Furthermore, these methods can only be applied a template-specific surface mesh, and not applicable more general meshes, like tetrahedrons non-manifold meshes. While graph convolution employed, lack performance in reconstruction...
Hands are dexterous and highly versatile manipulators that central to how humans interact with objects their environment. Consequently, modeling realistic hand-object interactions, including the subtle motion of individual fingers, is critical for applications in computer graphics, vision, mixed reality. Prior work on capturing interacting 3D focuses body object motion, often ignoring hand pose. In contrast, we introduce GRIP, a learning-based method takes, as input, object, synthesizes both...
Early retinal vascular changes in diseases such as diabetic retinopathy often occur at a microscopic level. Accurate evaluation of networks micro-level could significantly improve our understanding angiopathology and potentially aid ophthalmologists disease assessment management. Multiple angiogram-related imaging modalities, including fundus, optical coherence tomography angiography, fluorescence project continuous, inter-connected microvascular into domains. However, extracting the...
A robust pose estimation approach is proposed by combining facial appearance asymmetry and 3D geometry in a coarse-to-fine framework. The rough face first estimated analyzing the of distribution component detection confidences on an image, which actually implies intrinsic relation between appearance. Then, this pose, as well error bandwidth, utilized into 3D-to-2D geometrical model matching to refine estimation. able track with fast motion front cluttered background recover its robustly...
Clothes undergo complex geometric deformations, which lead to appearance changes. To edit human videos in a physically plausible way, texture map must take into account not only the garment transformation induced by body movements and clothes fitting, but also its 3D fine-grained surface geometry. This poses, however, new challenge of reconstruction dynamic from an image or video. In this paper, we show that it is possible dressed images without reconstruction. We estimate geometry aware...
For multi-view face alignment, we have to deal with two major problems: 1) the problem of multi-modality caused by diverse shape variation when view changes dramatically; 2) varying number feature points self-occlusion. Previous works used nonlinear models or based methods for alignment. However, they either assume all are visible apply a set discrete separately without uniform criterion. In this paper, propose unified framework solve in which, both and variable modeled Bayesian mixture...
Hyper-reflective foci (HRF) refers to the spot-shaped, block-shaped areas with characteristics of high local contrast and reflectivity, which is mostly observed in retinal optical coherence tomography (OCT) images patients fundus diseases. HRF mainly appears hard exudates (HE) microglia (MG) clinically. Accurate segmentation HE MG essential alleviate harm However, it still a challenge segment simultaneously due similar pathological features, various shapes location distribution, blurred...