A5062256552- Global Trade and Competitiveness
- Environmental Sustainability in Business
- Organic Food and Agriculture
- Global trade, sustainability, and social impact
- Generative Adversarial Networks and Image Synthesis
- Economic and Business Development Strategies
- Food Waste Reduction and Sustainability
- Culinary Culture and Tourism
- Environmental Education and Sustainability
- Regional Development and Management Studies
- Consumer Behavior in Brand Consumption and Identification
- Industrial Vision Systems and Defect Detection
- Image Processing Techniques and Applications
- Digital Media Forensic Detection
- Advanced Image Processing Techniques
- Sensory Analysis and Statistical Methods
- Sustainable Supply Chain Management
- Adversarial Robustness in Machine Learning
- Quality and Management Systems
- Engineering and Environmental Studies
- Digital Imaging for Blood Diseases
- Gaussian Processes and Bayesian Inference
- Image and Video Stabilization
- Sustainable Industrial Ecology
- Fault Detection and Control Systems
University of Glasgow
2022-2025
Forschungszentrum Jülich
2023
University of National and World Economy
2011-2019
Medical University of Varna
2019
South East European Research Centre
2011
Abstract For several years, B ulgaria has been implementing systems for separate collection of packaging waste as elements environment policy, which aims to decrease the quantity municipal‐generated deposited in land. The effectiveness these is largely determined by consumers' recycling behaviour post‐socialist countries with emerging sustainable patterns. aim this article identify different segments among ulgarians based on their attitudes towards order highlight characteristics recyclers...
We present DiffInfinite, a hierarchical diffusion model that generates arbitrarily large histological images while preserving long-range correlation structural information. Our approach first synthetic segmentation masks, subsequently used as conditions for the high-fidelity generative process. The proposed sampling method can be scaled up to any desired image size only requiring small patches fast training. Moreover, it parallelized more efficiently than previous large-content generation...
Accurately detecting and classifying damage in analogue media such as paintings, photographs, textiles, mosaics, frescoes is essential for cultural heritage preservation. While machine learning models excel correcting global degradation if the operator known a priori, we show that they fail to predict where even after supervised training; thus, reliable detection remains challenge. We introduce DamBench, dataset diverse media, with over 11,000 annotations covering 15 types across various...
We present a latent diffusion model over 3D scenes, that can be trained using only 2D image data. To achieve this, we first design an autoencoder maps multi-view images to Gaussian splats, and simultaneously builds compressed representation of these splats. Then, train the space learn efficient generative model. This pipeline does not require object masks nor depths, is suitable for complex scenes with arbitrary camera positions. conduct careful experiments on two large-scale datasets...
Accurately detecting and classifying damage in analogue media such as paintings, photographs, textiles, mosaics, frescoes is essential for cultural heritage preservation. While machine learning models excel correcting degradation if the operator known a priori, we show that they fail to robustly predict where even after supervised training; thus, reliable detection remains challenge. Motivated by this, introduce ARTeFACT, dataset diverse types media, with over 11,000 annotations covering 15...
We propose an unsupervised image segmentation method using features from pre-trained text-to-image diffusion models. Inspired by classic spectral clustering approaches, we construct adjacency matrices self-attention layers between patches and recursively partition Normalised Cuts. A key insight is that probability distributions, which capture semantic relations patches, can be interpreted as a transition matrix for random walks across the image. leverage this first Random Walk Normalized...