A5090409177- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Multimodal Machine Learning Applications
- Remote-Sensing Image Classification
- Visual Attention and Saliency Detection
- Radiation Detection and Scintillator Technologies
- Robotics and Automated Systems
- Hydrology and Watershed Management Studies
- COVID-19 diagnosis using AI
- Virtual Reality Applications and Impacts
- Human Pose and Action Recognition
- Advanced Memory and Neural Computing
- Face recognition and analysis
- Particle Detector Development and Performance
- Image and Signal Denoising Methods
- Flood Risk Assessment and Management
- Body Image and Dysmorphia Studies
- Hydrology and Drought Analysis
- Advanced Image Fusion Techniques
National Technical University of Athens
2021-2024
Inria Rennes - Bretagne Atlantique Research Centre
2022
Athena Research and Innovation Center In Information Communication & Knowledge Technologies
2022
The past few years have seen an accelerating integration of deep learning (DL) techniques into various remote sensing (RS) applications, highlighting their power to adapt and achieving unprecedented advancements. In the present review, we provide exhaustive exploration DL approaches proposed specifically for spatial downscaling RS imagery. A key contribution our work is presentation major architectural components models, metrics, data sets available this task as well construction a compact...
Convolutional networks and vision transformers have different forms of pairwise interactions, pooling across layers at the end network. Does latter really need to be different? As a by-product pooling, provide spatial attention for free, but this is most often low quality unless self-supervised, which not well studied. Is supervision problem?In work, we develop generic framework then formulate number existing methods as instantiations. By discussing properties each group methods, derive...
Metric learning involves a discriminative representation such that embeddings of similar classes are encouraged to be close, while dissimilar pushed far apart. State-of-the-art methods focus mostly on sophisticated loss functions or mining strategies. On the one hand, metric losses consider two more examples at time. other modern data augmentation for classification The combination ideas is under-studied. In this work, we aim bridge gap and improve representations using mixup, which powerful...
In this paper, we discuss the drought-alert decision support system (DA-DSS), a solution developed by Small Medium Enterprise Amigo s.r.l., (Rome, Italy) and MaP Ltd., (Athens, Greece).within Horizon 2020 cross-sectoral project "cross-climate".DA-DSS is prototype of WebGIS application aimed to water utilities (WUs) in management drinking changing climate.DA-DSS combines climate data, infrastructures geospatial visualization tools.Climate data exploit seasonal forecasts provided EU Copernicus...
Augmented Reality or AR filters on selfies have become very popular social media platforms for a variety of applications, including marketing, entertainment and aesthetics. Given the wide adoption face importance faces in our structures relations, there is increased interest by scientific community to analyze impact such from psychological, artistic sociological perspective. However, are few quantitative analyses this area mainly due lack publicly available datasets facial images with...
Real-time analysis of Martian craters is crucial for mission-critical operations, including safe landings and geological exploration. This work leverages the latest breakthroughs on-the-edge crater detection aboard spacecraft. We rigorously benchmark several YOLO networks using a Mars dataset, analyzing their performance on embedded systems with focus optimization low-power devices. optimize this process new wave cost-effective, commercial-off-the-shelf-based smaller satellites....
This work introduces composed image retrieval to remote sensing. It allows query a large archive by examples alternated textual description, enriching the descriptive power over unimodal queries, either visual or textual. Various attributes can be modified part, such as shape, color, context. A novel method fusing image-to-image and text-to-image similarity is introduced. We demonstrate that vision-language model possesses sufficient no further learning step training data are necessary....
This work addresses composed image retrieval in the context of domain conversion, where content a query is retrieved specified by text. We show that strong vision-language model provides sufficient descriptive power without additional training. The mapped to text input space using textual inversion. Unlike common practice invert continuous tokens, we use discrete word via nearest-neighbor search vocabulary. With this inversion, softly across vocabulary and made more robust retrieval-based...
Convolutional networks and vision transformers have different forms of pairwise interactions, pooling across layers at the end network. Does latter really need to be different? As a by-product pooling, provide spatial attention for free, but this is most often low quality unless self-supervised, which not well studied. Is supervision problem? In work, we develop generic framework then formulate number existing methods as instantiations. By discussing properties each group methods, derive...
Transformers and masked language modeling are quickly being adopted explored in computer vision as transformers image (MIM). In this work, we argue that token masking differs from text, due to the amount correlation of tokens an image. particular, generate a challenging pretext task for MIM, advocate shift random informed masking. We develop exhibit idea context distillation-based where teacher transformer encoder generates attention map, which use guide student. thus introduce novel...