A5068625652- Molten salt chemistry and electrochemical processes
- Advanced Combustion Engine Technologies
- Generative Adversarial Networks and Image Synthesis
- Advanced Image Processing Techniques
- Combustion and flame dynamics
- Video Analysis and Summarization
- Multimodal Machine Learning Applications
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Ionic liquids properties and applications
- Advancements in Battery Materials
- Bauxite Residue and Utilization
- Combustion and Detonation Processes
- Extraction and Separation Processes
- Inorganic Fluorides and Related Compounds
- Metallurgical Processes and Thermodynamics
- Computer Graphics and Visualization Techniques
- Electrochemical Analysis and Applications
- Biodiesel Production and Applications
- Image and Signal Denoising Methods
- Handwritten Text Recognition Techniques
- Domain Adaptation and Few-Shot Learning
- Music and Audio Processing
- Image Retrieval and Classification Techniques
- Natural Language Processing Techniques
Huazhong University of Science and Technology
2015-2024
Adobe Systems (United States)
2015-2024
Northeastern University
2015-2024
Central Conservatory of Music
2024
Dalian University of Technology
2024
Advanced Micro Devices (United States)
2024
Universidad del Noreste
2009-2023
Shanghai Jiao Tong University
2011-2023
Shanghai First People's Hospital
2013-2023
Northeast Forestry University
2023
Automatically generating a natural language description of an image has attracted interests recently both because its importance in practical applications and it connects two major artificial intelligence fields: computer vision processing. Existing approaches are either top-down, which start from gist convert into words, or bottom-up, come up with words describing various aspects then combine them. In this paper, we propose new algorithm that combines through model semantic attention. Our...
In this paper, we propose a novel coupled dictionary training method for single image super-resolution based on patchwise sparse recovery, where the learned couple dictionaries relate low- and high-resolution patch spaces via representation. The learning process enforces that representation of low-resolution in terms can well reconstruct its underlying with highresolution space. We model problem as bilevel optimization problem, includes an 1-norm minimization constraints. Implicit...
Deep learning techniques have been successfully applied in many areas of computer vision, including low-level image restoration problems. For super-resolution, several models based on deep neural networks recently proposed and attained superior performance that overshadows all previous handcrafted models. The question then arises whether large-capacity data-driven become the dominant solution to ill-posed super-resolution problem. In this paper, we argue domain expertise represented by...
This paper reviews the 2nd NTIRE challenge on single image super-resolution (restoration of rich details in a low resolution image) with focus proposed solutions and results. The had 4 tracks. Track 1 employed standard bicubic downscaling setup, while Tracks 2, 3 realistic unknown downgrading operators simulating camera acquisition pipeline. were learnable through provided pairs high train images. tracks 145, 114, 101, 113 registered participants, resp., 31 teams competed final testing...
Universal style transfer aims to arbitrary visual styles content images. Existing feed-forward based methods, while enjoying the inference efficiency, are mainly limited by inability of generalizing unseen or compromised quality. In this paper, we present a simple yet effective method that tackles these limitations without training on any pre-defined styles. The key ingredient our is pair feature transforms, whitening and coloring, embedded an image reconstruction network. coloring...
Due to the significant information loss in low-resolution (LR) images, it has become extremely challenging further advance state-of-the-art of single image super-resolution (SISR). Reference-based (RefSR), on other hand, proven be promising recovering high-resolution (HR) details when a reference (Ref) with similar content as that LR input is given. However, quality RefSR can degrade severely Ref less similar. This paper aims unleash potential by leveraging more texture from images stronger...
In this report we demonstrate that with same parameters and computational budgets, models wider features before ReLU activation have significantly better performance for single image super-resolution (SISR). The resulted SR residual network has a slim identity mapping pathway (\(2\times\) to \(4\times\)) channels in each block. To further widen (\(6\times\) \(9\times\)) without overhead, introduce linear low-rank convolution into networks achieve even accuracy-efficiency tradeoffs. addition,...
In this work, we focus on the challenge of taking partial observations highly-stylized text and generalizing to generate unobserved glyphs in ornamented typeface. To a set multi-content images following consistent style from very few examples, propose an end-to-end stacked conditional GAN model considering content along channels network layers. Our proposed transfers given contents unseen ones, capturing highly stylized fonts found real-world such as those movie posters or infographics. We...
Single image super-resolution (SR) is an ill-posed problem, which tries to recover a high-resolution from its low-resolution observation. To regularize the solution of previous methods have focused on designing good priors for natural images, such as sparse representation, or directly learning large data set with models, deep neural networks. In this paper, we argue that domain expertise conventional coding model can be combined key ingredients achieve further improved results. We...
Recent progresses on deep discriminative and generative modeling have shown promising results texture synthesis. However, existing feed-forward based methods trade off generality for efficiency, which suffer from many issues, such as shortage of (i.e., build one network per texture), lack diversity always produce visually identical output) suboptimality generate less satisfying visual effects). In this work, we focus solving these issues improved We propose a enables efficient synthesis...
Video super-resolution (SR) aims to generate a high-resolution (HR) frame from multiple low-resolution (LR) frames in local temporal window. The inter-frame relation is as crucial the intra-frame spatial for tackling this problem. However, how utilize information efficiently and effectively remains challenging since complex motion difficult model can introduce adverse effects if not handled properly. We address problem two aspects. First, we propose adaptive neural network that adaptively...
Automatically generating a natural language description of an image has attracted interests recently both because its importance in practical applications and it connects two major artificial intelligence fields: computer vision processing. Existing approaches are either top-down, which start from gist convert into words, or bottom-up, come up with words describing various aspects then combine them. In this paper, we propose new algorithm that combines through model semantic attention. Our...
Building effective recommender systems for domains like fashion is challenging due to the high level of subjectivity and semantic complexity features involved (i.e., styles). Recent work has shown that approaches 'visual' recommendation (e.g. clothing, art, etc.) can be made more accurate by incorporating visual signals directly into objective, using 'off-the-shelf' feature representations derived from deep networks. Here, we seek extend this contribution showing performance significantly...
As two of the five traditional human senses (sight, hearing, taste, smell, and touch), vision sound are basic sources through which humans understand world. Often correlated during natural events, these modalities combine to jointly affect perception. In this paper, we pose task generating given visual input. Such capabilities could help enable applications in virtual reality (generating for scenes automatically) or provide additional accessibility images videos people with impairments. a...
Anaplasma species are obligate intracellular rickettsial pathogens that impact the health of humans and animals. Few studies have been carried out on infections in central southern China. This study was conducted to determine coinfection rates ovis, A. bovis, phagocytophilum from 262 field blood samples goats these regions. The average prevalences single infection were 15.3, 16.0, 6.1%, respectively. Coinfection ovis bovis dominant, with an rate 27.1%. 1.9% 4.2%. Three-pathogen found three...
Unsupervised Domain Adaptation (UDA) aims to transfer domain knowledge from existing well-defined tasks new ones where labels are unavailable. In the real-world applications, as (task) discrepancies usually uncontrollable, it is significantly motivated match feature distributions even if disparate. Additionally, no label available in target domain, how successfully adapt classifier source still remains an open question. this paper, we propose Re-weighted Adversarial Network (RAAN) reduce...
Deep learning techniques have been successfully applied in many areas of computer vision, including low-level image restoration problems. For super-resolution, several models based on deep neural networks recently proposed and attained superior performance that overshadows all previous handcrafted models. The question then arises whether large-capacity data-driven become the dominant solution to ill-posed super-resolution problem. In this paper, we argue domain expertise represented by...
Artistic text style transfer is the task of migrating from a source image to target create artistic typography. Recent methods have considered texture control enhance usability. However, controlling stylistic degree in terms shape deformation remains an important open challenge. In this paper, we present first network that allows for real-time crucial glyph through adjustable parameter. Our key contribution novel bidirectional matching framework establish effective glyph-style mapping at...
Single image super-resolution (SR) aims to estimate a high-resolution (HR) from lowresolution (LR) input. Image priors are commonly learned regularize the otherwise seriously ill-posed SR problem, either using external LR-HR pairs or internal similar patterns. We propose joint adaptively combine advantages of both and methods. define two loss functions sparse coding based examples, epitomic matching on as well corresponding adaptive weight automatically balance their contributions according...