A5101992408- Domain Adaptation and Few-Shot Learning
- Face and Expression Recognition
- Advanced Neural Network Applications
- Face recognition and analysis
- Histone Deacetylase Inhibitors Research
- Acute Lymphoblastic Leukemia research
- Chronic Myeloid Leukemia Treatments
- Adversarial Robustness in Machine Learning
- Sparse and Compressive Sensing Techniques
- Advanced Image and Video Retrieval Techniques
- Privacy-Preserving Technologies in Data
- Multimodal Machine Learning Applications
- Advanced Vision and Imaging
- Human Pose and Action Recognition
- COVID-19 diagnosis using AI
- Image Processing Techniques and Applications
- Cell Image Analysis Techniques
- Image and Signal Denoising Methods
- Gaze Tracking and Assistive Technology
- Machine Learning and Data Classification
- Robotics and Sensor-Based Localization
- Advanced Image Processing Techniques
- Biometric Identification and Security
- Advanced Graph Neural Networks
- Child Development and Digital Technology
Purdue University West Lafayette
2019-2025
Nanjing Forestry University
2024
VA Maryland Health Care System
2024
Chinese Academy of Sciences
2006-2023
State Key Laboratory of Biotherapy
2023
Sichuan University
2023
Institute of Computing Technology
2017-2023
West China Hospital of Sichuan University
2023
RELX Group (United States)
2022
Duke University
2013-2021
Weakly supervised instance segmentation with image-level labels, instead of expensive pixel-level masks, remains unexplored. In this paper, we tackle challenging problem by exploiting class peak responses to enable a classification network for mask extraction. With image labels supervision only, CNN classifiers in fully convolutional manner can produce response maps, which specify confidence at each location. We observed that local maximums, i.e., peaks, map typically correspond strong...
Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling significant local and global rotations remains limited. In this paper, we propose Active Rotating Filters (ARFs) that actively rotate during convolution produce feature maps with location orientation explicitly encoded. An ARF acts as a virtual filter bank containing the itself its multiple unmaterialised rotated versions. During...
Domain adaptation addresses the problem where data instances of a source domain have different distributions from that target domain, which occurs frequently in many real life scenarios. This work focuses on unsupervised adaptation, labeled are only available domain. We propose to interpolate subspaces through dictionary learning link and domains. These able capture intrinsic shift form shared feature representation for cross recognition. Further, we introduce quantitative measure...
Weakly supervised object localization remains challenging, where only image labels instead of bounding boxes are available during training. Object proposal is an effective component in localization, but often computationally expensive and incapable joint optimization with some the remaining modules. In this paper, to best our knowledge, we for first time integrate weakly into convolutional neural networks (CNNs) end-to-end learning manner. We design a network component, Soft Proposal (SP),...
Lane detection is to detect lanes on the road and provide accurate location shape of each lane. It severs as one key techniques enable modern assisted autonomous driving systems. However, several unique properties challenge methods. The lack distinctive features makes lane algorithms tend be confused by other objects with similar local appearance. Moreover, inconsistent number a well diverse line patterns, e.g. solid, broken, single, double, merging, splitting lines further hamper...
Surveillance cameras today often capture NIR (near infrared) images in low-light environments. However, most face datasets accessible for training and verification are only collected the VIS (visible light) spectrum. It remains a challenging problem to match due different light Recently, breakthroughs have been made recognition by applying deep learning on huge amount of labeled samples. The same approach cannot be simply applied two main reasons: First, much limited available compared...
To demonstrate the capability of computer vision analysis to detect atypical orienting and attention behaviors in toddlers with autism spectrum disorder. One hundered four 16-31 months old (mean = 22) participated this study. Twenty-two had disorder 82 typical development or developmental delay. Toddlers watched video stimuli on a tablet while built-in camera recorded their head movement. Computer measured participants' response name calls. Reliability algorithm was tested against human...
Current tools for objectively measuring young children's observed behaviors are expensive, time-consuming, and require extensive training professional administration. The lack of scalable, reliable, validated impacts access to evidence-based knowledge limits our capacity collect population-level data in non-clinical settings. To address this gap, we developed mobile technology videos children while they watched movies designed elicit autism-related then used automatic behavioral coding these...
Although facial expressions can be decomposed in terms of action units (AUs) as suggested by the coding system, there have been only a few attempts that recognize expression using AUs and their composition rules. In this paper, we propose dictionary-based approach for analysis decomposing AUs. First, construct an AU-dictionary domain experts' knowledge To incorporate high-level regarding decomposition AUs, then perform structure-preserving sparse imposing two layers grouping over atoms well...
Observational behavior analysis plays a key role for the discovery and evaluation of risk markers many neurodevelopmental disorders. Research on autism spectrum disorder (ASD) suggests that behavioral can be observed at 12 months age or earlier, with diagnosis possible 18 months. To date, these studies evaluations involving observational tend to rely heavily clinical practitioners specialists who have undergone intensive training able reliably administer carefully designed...
Deep neural networks trained using a softmax layer at the top and cross-entropy loss are ubiquitous tools for image classification. Yet, this does not naturally enforce intra-class similarity nor inter-class margin of learned deep representations. To simultaneously achieve these two goals, different solutions have been proposed in literature, such as pairwise or triplet losses. However, carry extra task selecting pairs triplets, computational burden computing learning many combinations them....
We present a two-stage approach for learning dictionaries object classification tasks based on the principle of information maximization. The proposed method seeks dictionary that is compact, discriminative, and generative. In first stage, atoms are selected from an initial by maximizing mutual measure compactness, discrimination reconstruction. second updated improved reconstructive discriminative power using simple gradient ascent algorithm information. Experiments real data sets...
Data-free Knowledge Distillation (DFKD) has gained popularity recently, with the fundamental idea of carrying out knowledge transfer from a Teacher neural network to Student in absence training data. However, Adversarial DFKD framework, student network's accuracy, suffers due non-stationary distribution pseudo-samples under multiple generator updates. To this end, at every update, we aim maintain student's performance on previously encountered examples while acquiring samples current...
LiDAR-based 3D object detection plays a crucial role in modern autonomous driving systems. LiDAR data often exhibit severe changes properties across different observation ranges. In this paper, we explore cross-range adaptation for using LiDAR, i.e., far-range observations are adapted to near-range. This way, is optimized similar performance near-range one. We adopt bird-eyes view (BEV) framework perform the proposed model adaptation. Our consists of an adversarial global adaptation, and...
Images captured in severe weather such as rain and snow significantly degrade the accuracy of vision systems, e.g., for outdoor video surveillance or autonomous driving. Image deraining is a critical yet highly challenging task, due to fact that density varies across spatial locations, while distribution patterns simultaneously vary color channels. In this paper, we propose variational image (VID) method by formulating conditional auto-encoder framework. To achieve adaptive density, generate...
Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling significant local and global rotations remains limited. In this paper, we propose Active Rotating Filters (ARFs) that actively rotate during convolution produce feature maps with location orientation explicitly encoded. An ARF acts as a virtual filter bank containing the itself its multiple unmaterialised rotated versions. During...