• Digital Humanities and Cultural Heritage
• 2012-2021close
• Publications close
• FI close
• NEANIAS Space Research Community close

A pioneering study showed that the fine structure in the luminosity function (LF) of young star clusters contains information about the evolutionary stage (age) and composition of the stellar population. The notable features include the H-peak, which is the result of the onset of hydrogen burning turning pre-main sequence stars into main sequence stars. The feature moves toward the faint end of the LF, and eventually disappears as the population evolves. Another detectable feature is the Wielen dip, a dip at M_V ~ 7 mag in the LF first identified in 1974 for stars in the solar environment. Later studies also identified this feature in the LF of star clusters. The Wielen dip is caused by the increased importance of H- opacity in a certain range of low-mass stars. We studied the detailed structure in the luminosity function using the data from Gaia DR2 and PARSEC stellar evolution models with the aim to further our understanding of young stellar populations. We analyzed the astrometric properties of stars in the solar neighborhood (< 20 pc) and in various relatively nearby (< 400 pc) young (< 50 Myr) open clusters and OB associations, and compare the features in the luminosity function with those generated by PARSEC models. The Wielen dip is confirmed in the LF of all the populations, including the solar neighborhood, at M_G ~7 mag. The H-peak is present in the LF of the field stars in the solar neighborhood. It likely signals that the population is mixed with a significant number of stars younger than 100 Myr. The H-peak is found in the LF of young open clusters and OB associations, and its location varies with age. Our observations with Gaia DR2 confirm the evolution of the H-peak from 5 Myr up to 47 Myr. The fine structure in the luminosity function in young stellar populations can be used to estimate their age. 16 pages, A&A in press

This paper introduces a design framework for mixed reality urban exploration (MRUE), based on a concrete implementation in a historical city. The framework integrates different modalities, such as virtual reality (VR), augmented reality (AR), and haptics-audio interfaces, as well as advanced features such as personalized recommendations, social exploration, and itinerary management. It permits to address a number of concerns regarding information overload, safety, and quality of the experience, which are not sufficiently tackled in traditional non-integrated approaches. This study presents an integrated mobile platform built on top of this framework and reflects on the lessons learned. Funding Information: Acknowledgments. This work was supported by the European Union Seventh Framework Programme (grant agreement no 601139) and the Italian Ministry of Education, University, and Research (MIUR) PON AIM project (id: AIM1875400-1, CUP: B74I18000210006). Publisher Copyright: © 2021. All Rights Reserved. Peer reviewed

Labeling videos at scale is impractical. Consequently, self-supervised visual representation learning is key for efficient video analysis. Recent success in learning image representations suggest contrastive learning is a promising framework to tackle this challenge. However, when applied to real-world videos, contrastive learning may unknowingly lead to separation of instances that contain semantically similar events. In our work, we introduce a cooperative variant of contrastive learning to utilize complementary information across views and address this issue. We use data-driven sampling to leverage implicit relationships between multiple input video views, whether observed (e.g. RGB) or inferred (e.g. flow, segmentation masks, poses). We are one of the firsts to explore exploiting inter-instance relationships to drive learning. We experimentally evaluate our representations on the downstream task of action recognition. Our method achieves competitive performance on standard benchmarks (UCF101, HMDB51, Kinetics400). Furthermore, qualitative experiments illustrate that our models can capture higher-order class relationships. The code is available at http://github.com/nishantrai18/CoCon.

Abstract Construction of managed aquifer recharge structures (MARS) to store floodwater is a common strategy for storing depleted groundwater resources in arid and semi-arid regions, as part of integrated water resources management (IWRM). MARS divert surface water to groundwater, but this can affect downstream fluvial processes. The impact of MARS on fluvial processes was investigated in this study by combining remote sensing techniques with hydro-sediment modeling for the case of the Kaboutar-Ali-Chay aquifer, northwestern Iran. The impact of MARS on groundwater dynamics was assessed, sedimentation across the MARS was modeled using a 2D hydrodynamic model, and morphological changes were quantified in the human-impacted alluvial fan using Landsat time series data and statistical methods. Changes were detected by comparing data for the periods before (1985–1996) and after (1997–2018) MARS construction. The results showed that the rate of groundwater depletion decreased from 2.14 m/yr before to 0.86 m/yr after MARS construction. Hydro-sediment modeling revealed that MARS ponds slowed water outflow, resulting in a severe decrease in sediment load which lead to a change from sediment deposition to sediment erosion in the alluvial fan. Morphometric analyses revealed decreasing alluvial fan area and indicated significant differences (p < 0.01) between pre- and post-impact periods for different morphometric parameters analyzed. The rate of change in area of the Kaboutar-Ali-Chay alluvial fan changed from −0.228 to −0.115 km²/year between pre- and post-impact periods.

We develop an approach to learning visual representations that embraces multimodal data, driven by a combination of intra- and inter-modal similarity preservation objectives. Unlike existing visual pre-training methods, which solve a proxy prediction task in a single domain, our method exploits intrinsic data properties within each modality and semantic information from cross-modal correlation simultaneously, hence improving the quality of learned visual representations. By including multimodal training in a unified framework with different types of contrastive losses, our method can learn more powerful and generic visual features. We first train our model on COCO and evaluate the learned visual representations on various downstream tasks including image classification, object detection, and instance segmentation. For example, the visual representations pre-trained on COCO by our method achieve state-of-the-art top-1 validation accuracy of $55.3\%$ on ImageNet classification, under the common transfer protocol. We also evaluate our method on the large-scale Stock images dataset and show its effectiveness on multi-label image tagging, and cross-modal retrieval tasks.

Generalization to out-of-distribution data has been a problem for Visual Question Answering (VQA) models. To measure generalization to novel questions, we propose to separate them into "skills" and "concepts". "Skills" are visual tasks, such as counting or attribute recognition, and are applied to "concepts" mentioned in the question, such as objects and people. VQA methods should be able to compose skills and concepts in novel ways, regardless of whether the specific composition has been seen in training, yet we demonstrate that existing models have much to improve upon towards handling new compositions. We present a novel method for learning to compose skills and concepts that separates these two factors implicitly within a model by learning grounded concept representations and disentangling the encoding of skills from that of concepts. We enforce these properties with a novel contrastive learning procedure that does not rely on external annotations and can be learned from unlabeled image-question pairs. Experiments demonstrate the effectiveness of our approach for improving compositional and grounding performance. Comment: Paper at CVPR 2021. 14 pages, 7 figures

Conducting archaeological site surveys is time consuming, and large sites may have many small features or structures that are difficult to locate and interpret. Vegetation cover and dense forest hide small structures, like cairns, while at the same time forest cover can cause problems for LiDAR tools. In this case study, drone-based ALS (airborne laser scanning) was tested as an archaeological site survey tool. The research site was complex and located partially in a forested area, which made it possible to evaluate how forest cover affects data. The survey methods used were rather simple: visual analysis, point density calculations in the forest area, and, for site interpretation purposes, digitizing observations and viewshed analysis. Using straightforward methods allowed us to evaluate the minimum time and skills needed for this type of survey. Drone-based ALS provided good results and increased knowledge of the site and its structures. Estimates of the number of cairns interpreted as graves more than doubled as a result of the high-accuracy ALS data. Based on the results of this study, drone-based ALS could be a suitable high-accuracy survey method for large archaeological sites. However, forest cover affects the accuracy, and more research is needed. Peer reviewed