Abstract

Abstract

Abstract
We present outcomes from our experimental work towards identification of forest segments in Czech Jeseniky mountains damaged by a hurricane event on March 17, 2018. We have specifically processed Sentinel-1 satellite radar data and identified a functional methodology of extracting extents of the affected segments. The backscatter intensity of the damaged forest segments in Sentinel-1 images does not change significantly, subject to the sensitivity of the instrument. We have identified that a careful preprocessing of the data can lead to a state of possibility to identify edges of the affected areas in one of Principal Components (PC) generated from a set of dual-polarisation images before and after the event. In our case, these features were clearly visible in PC3 that was used in post-processing chain incorporating strong spatial filtering and edge detection routines. The identified damaged forest segments were validated by mapping during visiting one of the areas and by a comparison with multispectral satellite imagery, from data taken following year (as the damaged forest areas were already cleared and not regenerated). The approach can bring advantage in possibility of early preliminary mapping of the forest damages. (C) 2021 The Authors. Published by Elsevier B.V.

Abstract
Authoring 360 multisensory videos is a true challenge as the authoring tools available are scarce and restrictive. In this paper, we propose an authoring tool with three different authoring interfaces (desktop, immersive, and tangible interface) for creating multisensory 360 videos with the advantage of having a live preview of the multisensory content that is being produced. An evaluation of the three authoring tools having into account gender, system usability, presence, satisfaction, and effectiveness (time to accomplish tasks, number of errors, and number of help requests) is presented. The sample consisted of 48 participants (24 males and 24 females) evenly distributed between the different interfaces (8 males and 8 females for each interface). The results revealed that gender does not have any impact in the studied interfaces regarding all the dependent variables; immersive and tangible interfaces have higher levels of satisfaction than desktop interface as it allows more interaction freedom, and desktop interface have the lowest time to accomplish the tasks because people are more familiar with keyboard and mouse.

Abstract
Multi-view has the potential to offer immersive viewing experiences to users, as an alternative to 360 degrees and Virtual Reality (VR) applications. In multi-view, a limited number of camera views are sent to the client and missing views are synthesised locally. Given the substantial complexity associated to view synthesis, considerable attention has been given to optimise the trade-off between bandwidth gains and computing resources, targeting smooth navigation and viewing quality. A still relatively unexplored field is the optimisation of the way navigation interactivity is achieved, i.e. how the user indicates to the system the selection of new viewpoints. In this article, we introduce SmoothMV, a multi-view system that uses a non-intrusive head tracking approach to enhance navigation and Quality of Experience (QoE) of the viewer. It relies on a novel Hot&Cold matrix concept to translate head positioning data into viewing angle selections. Streaming of selected views is done using MPEG-DASH, where a proposed extension to the standard descriptors enables to achieve consistent and flexible view identification.

Abstract
Due to convenience and usability, many deep learning (DL) jobs resort to the available shared parallel file system (PFS) for storing and accessing training data when running in HPC environments. Under such a scenario, however, where multiple I/O-intensive applications operate concurrently, the PFS can quickly get saturated with simultaneous storage requests and become a critical performance bottleneck, leading to throughput variability and performance loss. We present MONARCH, a framework-agnostic middleware for hierarchical storage management. This solution leverages the existing storage tiers present at modern supercomputers (e.g., compute node's local storage, PFS) to improve DL training performance and alleviate the current I/O pressure of the shared PFS. We validate the applicability of our approach by developing and integrating an early prototype with the TensorFlow DL framework. Results show that MONARCH can reduce I/O operations submitted to the shared PFS by up to 45%, decreasing training time by 24% and 12%, for I/O-intensive models, namely LeNet and AlexNet.