A5065955200- Advanced Vision and Imaging
- Image and Video Quality Assessment
- Video Coding and Compression Technologies
- Image Enhancement Techniques
- Human Pose and Action Recognition
- Computer Graphics and Visualization Techniques
- 3D Surveying and Cultural Heritage
- 3D Shape Modeling and Analysis
- Virtual Reality Applications and Impacts
- Robotics and Sensor-Based Localization
- Optical measurement and interference techniques
- Hand Gesture Recognition Systems
- Visual Attention and Saliency Detection
- Augmented Reality Applications
- Advanced Image Processing Techniques
- Generative Adversarial Networks and Image Synthesis
- Neural dynamics and brain function
- Advanced Neural Network Applications
- CCD and CMOS Imaging Sensors
- Robot Manipulation and Learning
- Explainable Artificial Intelligence (XAI)
- Telecommunications and Broadcasting Technologies
- Machine Learning in Materials Science
- Multimedia Communication and Technology
- Cell Image Analysis Techniques
Information Technologies Institute
2015-2024
Centre for Research and Technology Hellas
2018-2024
Maastricht University
2023-2024
China Philanthropy Research Institute
2020-2023
Hella (Germany)
2018
Information Technology Institute
2013-2017
During the last years, Convolutional Neural Networks (CNNs) have achieved state-of-the-art performance in image classification. Their architectures largely drawn inspiration by models of primate visual system. However, while recent research results neuroscience prove existence non-linear operations response complex cells, little effort has been devoted to extend convolution technique forms. Typical convolutional layers are linear systems, hence their expressiveness is limited. To overcome...
Depth perception is considered an invaluable source of information for various vision tasks. However, depth maps acquired using consumer-level sensors still suffer from non-negligible noise. This fact has recently motivated researchers to exploit traditional filters, as well the deep learning paradigm, in order suppress aforementioned non-uniform noise, while preserving geometric details. Despite effort, denoising open challenge mainly due lack clean data that could be used ground truth. In...
Recent advances in media capture and processing technologies have enabled new forms of true 3-D content that increase the degree user immersion. The demand for more engaging entertainment means distributors broadcasters need to fine-tune their delivery mechanisms over Internet as well develop models quantifying predicting experience these content. In work described this paper, we undertake one first studies into quality (QoE) real-time streamed virtual reality (VR) headsets purposes, context...
In this paper, a novel skeleton-based approach to human time-varying mesh (H-TVM) compression is presented. The topic of TVM new and has many challenges, such as handling the lack obvious mapping vertices across frames variable connectivity frames, while maintaining efficiency, which are most important ones. Very few works exist in literature, not all challenges have been addressed yet. addition, developing an efficient real-time solution, above, obviously difficult task. We attempt address...
Tele-immersion (TI) related technologies can change the way people interact and bridge gap between physical digital worlds. However, while technology itself advances, most developed platforms have complex setups require large investments. In this work, a low-cost platform is introduced, integrating multiple TI-related advances. Focusing on ease of use rapid deployment, fast fully automatic calibration method proposed. The enables real-time 3D reconstruction users their placement into...
This paper provides a systematic understanding of the requirements live 3D mesh coding, targeting (tele-)immersive media streaming applications. We thoroughly benchmark in rate-distortion and runtime performance terms, four static coding solutions that are openly available. Apart from geometry connectivity, our analysis includes experiments for compressing vertex normals attributes, something scarcely found literature. In addition, we provide theoretical model tele-immersion pipeline...
Multi-view capture systems are complex to engineer. They require technical knowledge install and intricate processes setup related mainly the sensors' spatial alignment (i.e. external calibration). However, with ongoing developments in new production methods, we now at a position where of high quality realistic 3D assets is possible even commodity sensors. Nonetheless, capturing developed these methods heavily intertwined themselves, relying on custom solutions seldom - if not all publicly...
Hereby, a new publicly available 3D reconstruction-oriented dataset is presented. It consists of multi-view range scans small-sized objects using turntable. Range were captured Microsoft Kinect sensor, as well an accurate laser scanner (Vivid VI-700 Non-contact Digitizer), whose reconstructions can serve ground-truth data. The construction this was motivated by the lack relevant dataset, despite fact that has attracted attention many researchers and home enthusiasts. Thus, core idea behind...
In this work, we explore the potential of exploiting activity-related global features in order to improve performance an existing human Time-Varying Mesh (TVM) compression scheme. The TVM scheme used, employs two kinds frames, namely Intra(I)-Fames and Enhanced Predicted(EP) Frames. scheme, I-Frames are used as a reference encode EP-Frames. paper introduces strategy for selecting most appropriate I-Frame that will serve frame encoding EP-Frames, characteristics. Two different strategies...
The upcoming 5G networks, among other technological advances, bring Network Function Virtualization (NFV) capabilities enabling deployment of application service intelligence on their Next Generation Core (NGC). Application specific logic is packaged into Virtual Functions (VNFs) so that instantiation and can be done at any node the NGC, with management orchestration being maintained by infrastructure. While number instances each VNF placement inside NGC network are managed infrastructure,...
An important line of research attempts to explain CNN image classifier predictions and intermediate layer representations in terms human understandable concepts. In this work, we expand on previous works the literature that use annotated concept datasets extract interpretable feature space directions propose an unsupervised post-hoc method a disentangling basis by looking for rotation explains sparse one-hot thresholded transformed pixel activations. We do experimentation with existing...
With the advent of consumer grade depth sensors, low-cost volumetric capture systems are easier to deploy. Their wider adoption though depends on their usability and by extension practicality spatially aligning multiple sensors. Most existing alignment approaches employ visual patterns, e.g. checkerboards, or markers require high user involvement technical knowledge. More user-friendly easier-to-use rely markerless methods that exploit geometric patterns a physical structure. However,...
Network functions virtualization (NFV) attributes to the substitute of network on dedicated appliances such as load balancers and routers with use virtualized instances running software. Any enterprise can easily implement a wide array using NFV while maximizing efficiencies introducing new revenue-generating services that are significantly faster easier than ever before. is key enabler coming 5G infrastructure, supporting various in network. The scope this document provide comprehensive...
Recent advances in full body 3D reconstruction methods have lead to the realisation of high quality, real-time, photo realistic capture users a range tele-immersion (TI) contexts including gaming and mixed reality environments. The (FBR) process is computationally expensive requiring comparatively CPU, GPU network resources order maintain shared, virtual which quality reproductions can be rendered real-time. A significant optimisation delivery FBR content has been achieved through real-time...
The task of transforming a furnished room image into background-only is extremely challenging since it requires making large changes regarding the scene context while still preserving overall layout and style. In order to acquire photo-realistic structural consistent background, existing deep learning methods either employ inpainting approaches or incorporate as an individual leverage later in not fully differentiable semantic region-adaptive normalization module. To tackle these drawbacks,...
With the advent of consumer grade depth sensors, low-cost volumetric capture systems are easier to deploy. Their wider adoption though depends on their usability and by extension practicality spatially aligning multiple sensors. Most existing alignment approaches employ visual patterns, e.g. checkerboards, or markers require high user involvement technical knowledge. More user-friendly easier-to-use rely markerless methods that exploit geometric patterns a physical structure. However,...