A5101923745- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Image Processing Techniques and Applications
- Image and Signal Denoising Methods
- Image Enhancement Techniques
- Generative Adversarial Networks and Image Synthesis
- Air Quality Monitoring and Forecasting
- Neural Networks and Applications
- 3D Shape Modeling and Analysis
- Humor Studies and Applications
- Music and Audio Processing
- 3D Surveying and Cultural Heritage
- Image Retrieval and Classification Techniques
- Multi-Criteria Decision Making
- Fuzzy Logic and Control Systems
- Color Science and Applications
- Advanced Image Fusion Techniques
- Video Surveillance and Tracking Methods
- Fuzzy Systems and Optimization
- Optical measurement and interference techniques
- Video Analysis and Summarization
- Machine Learning and Data Classification
- Advanced Optical Sensing Technologies
- Text and Document Classification Technologies
- Aerospace Engineering and Control Systems
Lappeenranta-Lahti University of Technology
2023-2025
Zhejiang Ocean University
2016-2024
Saint Francis University
2021-2023
DELL (United States)
2023
Icahn School of Medicine at Mount Sinai
1997-2022
Laboratoire d'Informatique de l'École Polytechnique
2020-2022
Centre National de la Recherche Scientifique
2020-2022
Hong Kong Polytechnic University
2015-2021
École Polytechnique
2020
Peking University
2020
Low-light image enhancement is a challenging task that has attracted considerable attention. Pictures taken in low-light conditions often have bad visual quality. To address the problem, we regard as residual learning problem to estimate between low- and normal-light images. In this paper, propose novel Deep Lightening Network (DLN) benefits from recent development of Convolutional Neural Networks (CNNs). The proposed DLN consists several Back-Projection (LBP) blocks. LBPs perform lightening...
This paper reviews the NTIRE 2020 challenge on real world super-resolution. It focuses participating methods and final results. The addresses setting, where paired true high low-resolution images are unavailable. For training, only one set of source input is therefore provided along with a unpaired high-quality target images. In Track 1: Image Processing artifacts, aim to super-resolve synthetically generated image processing artifacts. allows for quantitative benchmarking approaches w.r.t....
This paper reviewed the 3rd NTIRE challenge on single-image super-resolution (restoration of rich details in a low-resolution image) with focus proposed solutions and results. The had 1 track, which was aimed at real-world single image problem an unknown scaling factor. Participants were mapping images captured by DSLR camera shorter focal length to their high-resolution longer length. With this challenge, we introduced novel dataset (RealSR). track 403 registered participants, 36 teams...
Deep learning based image Super-Resolution (SR) has shown rapid development due to its ability of big data digestion. Generally, deeper and wider networks can extract richer feature maps generate SR images with remarkable quality. However, the more complex network we have, time consumption is required for practical applications. It important have a simplified efficient SR. In this paper, propose an Attention Back Projection Network (ABPN) super-resolution. Similar some recent works, believe...
This paper reviews the NTIRE 2021 challenge on learning super-Resolution space. It focuses participating methods and final results. The addresses problem of a model capable predicting space plausible super-resolution (SR) images, from single low-resolution image. must thus be sampling diverse outputs, rather than just generating SR goal is to spur research into developing formulations models better suited for highly ill-posed problem. And thereby advance state-of-the-art in broader field. In...
There is a great leap in objective accuracy on image super-resolution, which recently brings new challenge super-resolution with larger up-scaling (e.g. 4×) using pixel based distortion for measurement. This causes over-smooth effect cannot grasp well the perceptual similarity. The advent of generative adversarial networks makes it possible super-resolve low-resolution to generate photo-realistic images sharing distribution high-resolution images. However, suffer from problems mode-collapse...
Deep learning based single image super-resolution methods use a large number of training datasets and have recently achieved great quality progress both quantitatively qualitatively. Most deep networks focus on nonlinear mapping from low-resolution inputs to high-resolution outputs via residual without exploring the feature abstraction analysis. We propose Hierarchical Back Projection Network (HBPN), that cascades multiple HourGlass (HG) modules bottom-up top-down process features across all...
Abstract Automatically understanding funny moments (i.e., the that make people laugh) when watching comedy is challenging, as they relate to various features, such body language, dialogues and culture. In this paper, we propose FunnyNet-W, a model relies on cross- self-attention for visual, audio text data predict in videos. Unlike most methods rely ground truth form of subtitles, work exploit modalities come naturally with videos: (a) video frames contain visual information indispensable...
Diffusion models are a powerful framework for tackling ill-posed problems, with recent advancements extending their use to point cloud upsampling. Despite potential, existing diffusion struggle inefficiencies as they map Gaussian noise real clouds, overlooking the geometric information inherent in sparse clouds. To address these inefficiencies, we propose PUFM, flow matching approach directly clouds high-fidelity dense counterparts. Our method first employs midpoint interpolation resolving...
Air quality prediction is key to mitigating health impacts and guiding decisions, yet existing models tend focus on temporal trends while overlooking spatial generalization. We propose AQ-Net, a spatiotemporal reanalysis model for both observed unobserved stations in the near future. AQ-Net utilizes LSTM multi-head attention regression. also cyclic encoding technique ensure continuous time representation. To learn fine-grained air estimation, we incorporate with neural kNN explore...
Face hallucination or super-resolution is a practical application of general image which has been recently studied by many researchers. The challenge good face comes from variety poses, illuminations, facial expressions, and other degradations. In proposed methods, researchers resolve it using generative neural network to reduce the perceptual loss so we can generate photo-realistic image. problem that usually overlook fidelity super-resolved could affect further processing. Meanwhile, CNN...
This paper reviews the AIM 2019 challenge on extreme image super-resolution, problem of restoring rich details in a low resolution image. Compared to previous, this focuses an upscaling factor, ×16, and employs novel DIVerse 8K (DIV8K) dataset. report proposed solutions final results. The had 2 tracks. goal Track 1 was generate super-resolution result with high fidelity, using conventional PSNR as primary metric evaluate different methods. instead focused generating visually more pleasant...
In this paper, we propose a novel reference based image super-resolution approach via Variational AutoEncoder (RefVAE). Existing state-of-the-art methods mainly focus on single which cannot perform well large upsampling factors, e.g., 8×. We super-resolution, for any arbitrary can act as super-resolution. Even using random map or low-resolution itself, the proposed RefVAE transfer knowledge from to super-resolved images. Depending upon different references, method generate versions of images...
Image style transfer has attracted widespread attention in the past few years. Despite its remarkable results, it requires additional images available as references, making less flexible and inconvenient. Using text is most natural way to describe style. More importantly, can implicit abstract styles, like styles of specific artists or art movements. In this paper, we propose a text-driven image (TxST) that leverages advanced image-text encoders control arbitrary transfer. We introduce...
Modern works on style transfer focus transferring from a single image. Recently, some approaches study multiple transfer; these, however, are either too slow or fail to mix styles. We propose ST-VAE, Variational AutoEncoder for latent space-based transfer. It performs by projecting nonlinear styles linear space, enabling merge via interpolation before the new content To evaluate we experiment COCO and also present case revealing that ST-VAE outperforms other methods while being faster,...
Image style transfer has attracted widespread attention in the past years. Despite its remarkable results, it requires additional images available as references, making less flexible and inconvenient. Using text is most natural way to describe style. Text can implicit abstract styles, like styles of specific artists or art movements. In this work, we propose a text-driven (TxST) that leverages advanced image-text encoders control arbitrary transfer. We introduce contrastive training strategy...
This paper proposes a novel learning-based image super-resolution via weighted random forest model (SWRF). The proposed method uses the LR-HR training data to train model. underlying idea of this approach is use several decision trees classify based on simple splitting threshold value at each class. A linear regression learnt map relationship between LR and HR patches. During up-sampling process, obtain more robust super-resolved image, instead averaging models from different trees, biased...
Due to the development of deep learning, image super- resolution has achieved huge improvement on both subjective and objective qualities. However, computation is still a problem for real-time applications. In this paper, we propose Cascaded Random Forest Image Super-Resolution (CRFSR) which screens sufficient simple features train much robust efficient model super-resolution. To further boost up super-resolution performance, an extra Gaussian Mixture Model (GMM) based layer added as final...
This paper proposes a novel learning-based image Super-Resolution via Randomized Multi-split Forests model (SRRMF). The proposed method uses the LR-HR training patch pairs to nonlinear manifold into of linear subspaces. key idea this approach is use several decision trees split randomly data different classes. A regression learnt map relationship between LR and HR patches at end leaf nodes. In order make full generalization ability random forests, we randomize grow tree cover more...
This paper introduces the real image Super-Resolution (SR) challenge that was part of Advances in Image Manipulation (AIM) workshop, held conjunction with ECCV 2020. involves three tracks to super-resolve an input for $\times$2, $\times$3 and $\times$4 scaling factors, respectively. The goal is attract more attention realistic degradation SR task, which much complicated challenging, contributes real-world super-resolution applications. 452 participants were registered total, 24 teams...