A5013682396- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Image Processing Techniques and Applications
- Image Enhancement Techniques
- 3D Shape Modeling and Analysis
- Generative Adversarial Networks and Image Synthesis
- Human Pose and Action Recognition
- Photoacoustic and Ultrasonic Imaging
- Computer Graphics and Visualization Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Human Motion and Animation
- Face recognition and analysis
- COVID-19 diagnosis using AI
- Advanced Optical Imaging Technologies
- Advanced Image and Video Retrieval Techniques
- Multimedia Communication and Technology
- Seismic Imaging and Inversion Techniques
- Image and Video Quality Assessment
- CCD and CMOS Imaging Sensors
- Video Surveillance and Tracking Methods
- Advanced X-ray Imaging Techniques
- Medical Imaging Techniques and Applications
- Time Series Analysis and Forecasting
Huawei Technologies (Sweden)
2021-2024
University of Würzburg
2023
Huawei Technologies (China)
2021-2023
Huawei Technologies (United Kingdom)
2022
Technicolor (Germany)
2013-2016
Leibniz University Hannover
2013-2016
This paper presents a fast, high-performance method for super resolution with external learning. The first contribution leading to the excellent performance is bimodal tree clustering, which successfully exploits antipodal invariance of coarse-to-high-res mapping natural image patches and provides scalability finer partitions underlying coarse patch space. During training an ensemble such trees computed, providing different linearizations mapping. second main fast inference algorithm,...
This paper reviews the first challenge on high-dynamic range (HDR) imaging that was part of New Trends in Image Restoration and Enhancement (NTIRE) workshop, held conjunction with CVPR 2021. manuscript focuses newly introduced dataset, proposed methods their results. The aims at estimating a HDR image from one or multiple respective low-dynamic (LDR) observations, which might suffer under-or over-exposed regions different sources noise. is composed by two tracks: In Track 1 only single LDR...
This paper reviews the challenge on constrained high dynamic range (HDR) imaging that was part of New Trends in Image Restoration and Enhancement (NTIRE) workshop, held conjunction with CVPR 2022. manuscript focuses competition set-up, datasets, proposed methods their results. The aims at estimating an HDR image from multiple respective low (LDR) observations, which might suffer under-or over-exposed regions different sources noise. is composed two tracks emphasis fidelity complexity...
The main challenge in Super Resolution (SR) is to discover the mapping between low-and high-resolution manifolds of image patches, a complex ill-posed problem which has recently been addressed through piecewise linear regression with promising results. In this paper we present novel regression-based SR algorithm that benefits from an extended knowledge structure both manifolds. We propose transform collapses 16 variations induced dihedral group transforms (i.e. rotations, vertical and...
Digital cameras transform sensor RAW readings into RGB images by means of their Image Signal Processor (ISP). Computational photography tasks such as image denoising and colour constancy are commonly performed in the domain, part due to inherent hardware design, but also appealing simplicity noise statistics that result from direct readings. Despite this, availability is limited comparison with abundance diversity available data. Recent approaches have attempted bridge this gap estimating...
High dynamic range (HDR) imaging is of fundamental importance in modern digital photography pipelines and used to produce a high-quality photograph with well exposed regions despite varying illumination across the image. This typically achieved by merging multiple low (LDR) images taken at different exposures. However, over-exposed misalignment errors due poorly compensated motion result artefacts such as ghosting. In this paper, we present new HDR technique that specifically models...
Hand-object interaction modeling from a single RGB image is significantly challenging task. Previous works typically reconstruct hand-object interactions as texture-less meshes, ignoring photo-realistic generation. In this work, we introduce the HO123, novel method to synthesize novel-view images image. To end, first train 2D diffusion prior. Given camera pose in views, our approach transfers information into explicit hand representations, including depth and skeleton images. We propose...
Dictionary-based super-resolution (SR) algorithms usually select dictionary atoms based on the distance or similarity metrics. Although optimal selection of nearest neighbors is central importance for such methods, impact using proper metrics SR has been overlooked in literature, mainly due to vast usage Euclidean distance. In this paper, we present a very fast regression-based algorithm, which builds densely populated anchored neighborhoods and sublinear search structures. We perform study...
Interactions between human and objects are influenced not only by the object's pose shape, but also physical attributes such as object mass surface friction. They introduce important motion nuances that essential for diversity realism. Despite advancements in recent kinematics-based methods, this aspect has been overlooked. Generating nuanced presents two challenges. First, it is non-trivial to learn from multi-modal information derived both non-physical attributes. Second, there exists no...
This paper presents a fast Super-Resolution (SR) algorithm based on selective patch processing. Motivated by the observation that some regions of images are smooth and unfocused can be properly upscaled with interpolation methods, we locally estimate probability performing degradation-free upscaling. Our proposed framework explores usage supervised machine learning techniques tackles problem using binary boosted tree classifiers. The applied upscaler is chosen obtained probabilities: (1) A...
This manuscript presents the results of "A View Synthesis Challenge for Humans Heads (VSCHH)", which was part ICCV 2023 workshops. paper describes competition setup and provides details on replicating our initial baseline, TensoRF. Additionally, we provide a summary participants' methods their in benchmark table. The challenge aimed to synthesize novel camera views human heads using given set sparse training view images. proposed solutions participants were evaluated ranked based objective...
Recent advances in neural radiance fields enable novel view synthesis of photo-realistic images dynamic settings, which can be applied to scenarios with human animation. Commonly used implicit backbones establish accurate models, however, require many input views and additional annotations such as masks, UV maps depth maps. In this work, we propose ParDy-Human (Parameterized Dynamic Human Avatar), a fully explicit approach construct digital avatar from little single monocular sequence....
Reconstruction of high-quality HDR images is at the core modern computational photography. Significant progress has been made with multi-frame reconstruction methods, producing high-resolution, rich and accurate color reconstructions high-frequency details. However, they are still prone to fail in dynamic or largely over-exposed scenes, where frame misalignment often results visible ghosting artifacts. Recent approaches attempt alleviate this by utilizing an event-based camera (EBC), which...
Training models continually to detect and classify objects, from new classes domains, remains an open problem. In this work, we conduct a thorough analysis of why how object detection forget catastrophically. We focus on distillation-based approaches in two-stage networks; the most-common strategy employed contemporary continual work.Distillation aims transfer knowledge model trained previous tasks -- teacher student while it learns task. show that works well for region proposal network, but...
We present a noise-aware single-image super-resolution (SI-SR) algorithm, which automatically cancels additive noise while adding detail learned from lower-resolution scales. In contrast with most SI-SR techniques, we do not assume the input image to be clean source of examples. Instead, adapt recent and efficient in-place cross-scale self-similarity prior for both learning fine examples reducing noise. Our experiments show promising performance, despite relatively simple algorithm. Both...
Recent research in piecewise linear regression for Super-Resolution has shown the positive impact of training regressors with densely populated clusters whose datapoints are tight Euclidean space. In this paper we further how to improve locality condition during and better select them testing time. We study characteristics metrics best suited algorithms, which comparisons usually made between normalized vectors that lie on unitary hypersphere. Even though distance been widely used purpose,...