A5101559354- Generative Adversarial Networks and Image Synthesis
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Advanced Image Processing Techniques
- Image Enhancement Techniques
- Adversarial Robustness in Machine Learning
- Machine Learning and Data Classification
- Advanced Vision and Imaging
- Machine Learning and ELM
- Anomaly Detection Techniques and Applications
- Neural Networks and Applications
- Computer Graphics and Visualization Techniques
- Advanced Multi-Objective Optimization Algorithms
- Data Stream Mining Techniques
- CCD and CMOS Imaging Sensors
- Privacy-Preserving Technologies in Data
- Mobile Ad Hoc Networks
- Metaheuristic Optimization Algorithms Research
- Face recognition and analysis
- Human Pose and Action Recognition
- Face and Expression Recognition
- Multimodal Machine Learning Applications
- Network Security and Intrusion Detection
- Image and Video Quality Assessment
- 3D Shape Modeling and Analysis
National University of Singapore
2022-2024
Yunnan University
2024
University of Würzburg
2023
Nanjing University
2020-2021
Baidu (China)
2021
Vision Technology (United States)
2021
Jiangxi University of Science and Technology
2006-2017
Zhejiang University
2016
Beijing University of Posts and Telecommunications
2013
Fast arbitrary neural style transfer has attracted widespread attention from academic, industrial and art communities due to its flexibility in enabling various applications. Existing solutions either attentively fuse deep feature into content without considering distributions, or adaptively normalize according the such that their global statistics are matched. Although effective, leaving shallow unexplored locally statistics, they prone unnatural output with unpleasing local distortions. To...
Neural painting refers to the procedure of producing a series strokes for given image and non-photo-realistically recreating it using neural networks. While reinforcement learning (RL) based agents can generate stroke sequence step by this task, is not easy train stable RL agent. On other hand, optimization methods search set parameters iteratively in large space; such low efficiency significantly limits their prevalence practicality. Different from previous methods, paper, we formulate task...
In this paper, we study \xw{dataset distillation (DD)}, from a novel perspective and introduce \emph{dataset factorization} approach, termed \emph{HaBa}, which is plug-and-play strategy portable to any existing DD baseline. Unlike conventional approaches that aim produce distilled representative samples, \emph{HaBa} explores decomposing dataset into two components: data \emph{Ha}llucination networks \emph{Ba}ses, where the latter fed former reconstruct image samples. The flexible...
Dataset distillation, also known as dataset condensation, aims to compress a large into compact synthetic one. Existing methods perform condensation by assuming fixed storage or transmission budget. When the budget changes, however, they have repeat synthesizing process with access original datasets, which is highly cumbersome if not infeasible at all. In this paper, we explore problem of slimmable extract smaller given only previous results. We first study limitations existing algorithms on...
This work reviews the results of NTIRE 2023 Challenge on Image Shadow Removal. The described set solutions were proposed for a novel dataset, which captures wide range object-light interactions. It consists 1200 roughly pixel aligned pairs real shadow free and affected images, captured in controlled environment. data was white-box setup, using professional equipment lights acquisition sensors. challenge had number 144 participants registered, out 19 teams compared final ranking. extend...
Vision Transformer has demonstrated impressive success across various vision tasks. However, its heavy computation cost, which grows quadratically with respect to the token sequence length, largely limits power in handling large feature maps. To alleviate previous works rely on either fine-grained self-attentions restricted local small regions, or global but shorten length resulting coarse granularity. In this paper, we propose a novel model, termed as Self-guided (SG-Former), towards...
We present the new Bokeh Effect Transformation Dataset (BETD), and review proposed solutions for this novel task at NTIRE 2023 Challenge. Recent advancements of mobile photography aim to reach visual quality full-frame cameras. Now, a goal in computational is optimize effect itself, which aesthetic blur out-of-focus areas an image. Photographers create by benefiting from lens optical properties.The work design neural network capable converting one another without harming sharp foreground...
Image relighting is attracting increasing interest due to its various applications. From a research perspective, im-age can be exploited conduct both image normalization for domain adaptation, and also data augmentation. It has multiple direct uses photo montage aesthetic enhancement. In this paper, we review the NTIRE 2021 depth guided challenge.We rely on VIDIT dataset each of our two challenge tracks, including information. The first track one-to-one where goal transform illumination...
Transformer-based models achieve favorable performance in artistic style transfer recently thanks to its global receptive field and powerful multi-head/layer attention operations. Nevertheless, the over-paramerized multi-layer structure increases parameters significantly thus presents a heavy burden for training. Moreover, task of transfer, vanilla Transformer that fuses content features by residual connections is prone content-wise distortion. In this paper, we devise novel model termed as...
In this paper, we study a novel task that enables partial knowledge transfer from pre-trained models, which term as Partial Network Cloning (PNC). Unlike prior methods update all or at least part of the parameters in target network throughout process, PNC conducts parametric "cloning" source and then injects cloned module to target, without modifying its parameters. Thanks transferred module, is expected gain additional functionality, such inference on new classes; whenever needed, can be...
Adversarial attacks constitute a notable threat to machine learning systems, given their potential induce erroneous predictions and classifications. However, within real-world contexts, the essential specifics of deployed model are frequently treated as black box, consequently mitigating vulnerability such attacks. Thus, enhancing transferability adversarial samples has become crucial area research, which heavily relies on selecting appropriate surrogate models. To address this challenge, we...
Machine learning society has witnessed the emergence of a myriad Out-of-Distribution (OoD) algorithms, which address distribution shift between training and testing by searching for unified predictor or invariant feature representation. However, task directly mitigating in unseen set is rarely investigated, due to unavailability during phase thus impossibility translator mapping distribution. In this paper, we explore how bypass requirement make translation useful OoD prediction. We propose...
Dataset distillation and dataset pruning are two prominent techniques for compressing datasets to improve computational storage efficiency. Despite their overlapping objectives, these approaches rarely compared directly. Even within each field, the evaluation protocols inconsistent across various methods, which complicates fair comparisons hinders reproducibility. Considering limitations, we introduce in this paper a benchmark that equitably evaluates methodologies both literatures. Notably,...
Social emotional optimisation algorithm (SEOA) is a newly developed evolutionary algorithm, which exhibits excellent performance for various engineering problems in real–world applications. However, SEOA may easily trap into local optima when solving complex multimodal function problems. This paper proposes novel social called GOSEOA, performs the generalised opposition–based learning (GOBL) strategy with certain probability during evolution process. The proposed uses to transform current...
Style transfer aims to render the style of a given image for reference another content reference, and has been widely adopted in artistic generation editing. Existing approaches either apply holistic global manner, or migrate local colors textures counterparts pre-defined way. In case, only one result can be generated specific pair images, which therefore lacks flexibility is hard satisfy different users with preferences. We propose here novel strategy termed Any-to-Any Transfer address this...
This study assesses the outcomes of NTIRE 2023 Challenge on Non-Homogeneous Dehazing, wherein novel techniques were proposed and evaluated new image dataset called HD-NH-HAZE. The HD-NH-HAZE contains 50 high resolution pairs real-life outdoor images featuring nonhomogeneous hazy corresponding haze-free same scene. haze was simulated using a professional setup that replicated real-world conditions scenarios. competition had 246 participants 17 teams submitted solutions for final testing...
This paper presents an high-throughput routing metric, which concentrates on optimizing the behavior of expected transmission count metric (ETX) in real-world multi-hop Ad hoc networks. In practical applications, ETX always fluctuates over time even fixed scenarios, and does not incorporate effects gray area fluctuating links. Considering these drawbacks, we devise average with threshold (AETXT). Compared to ETX, AETXT makes two improvements: 1) incorporating (AETX) instead as criterion link...
A variety of modern AI products essentially require raw user data for training diverse machine learning models. With the increasing concern on privacy, federated learning, a decentralized framework, enables privacy-preserving models by iteratively aggregating model updates from participants, instead data. Since all i.e., mobile devices, need to transfer their local concurrently and over edge networks, network is easily overloaded, leading high risk transmission failures. Although previous...
As an effective solution for deepwater oil/gas export, steel lazy wave riser (SLWR) could effectively accommodate large platform offsets, decouple motions from the touchdown point and reduce top tension in marine environment. Since SLWR dynamic response is very essential design phase fatigue analysis, it has already drawn extensive research attention. This paper puts forward a developed mathematical model on basis of rod theory finite element method to predict static/dynamic riser. Newmark-β...
As a very important research issue in digital media art, neural learning based video style transfer has attracted more and attention. A lot of recent works import optical flow method to original image framework preserve frame-coherency prevent flicker. However, these methods highly rely on paired datasets content stylized video, which are often difficult obtain. Another limitation existing is that while maintaining inter-frame coherency, they will introduce strong ghosting artifacts. In...
Gravitational search algorithm (GSA) is an emerging evolutionary (EA), which has exhibited remarkable performance in many applications. However, the traditional gravitational tends to yield slow convergence speed when facing some complicated real-life problems. Aiming at this weakness, a new with Gaussian mutation strategy (GMGSA) presented. At each generation, GMGSA calculates centre of current individual and global best individual, then combines obtained information into generate...