The integration of information acquired with different modalities, spatial resolution and spectral bands has shown to improve predictive accuracies. Data fusion is therefore one of the key challenges in remote sensing. Most prior work focusing on multi-modal fusion, assumes that modalities are always available during inference. This assumption limits the applications of multi-modal models since in practice the data collection process is likely to generate data with missing, incomplete or corrupted modalities. In this paper, we show that Generative Adversarial Networks can be effectively used to overcome the problems that arise when modalities are missing or incomplete. Focusing on semantic segmentation of building footprints with missing modalities, our approach achieves an improvement of about 2% on the Intersection over Union (IoU) against the same network that relies only on the available modality.
The progress in electroacoustics offers a miniaturized, head-related sound source to produce spatial auditory events [F. M. Koenig, AES Preprint No. 4495 (1997)]. These objects, so-called headphones, are working as dynamic (electromagnetic) or electrostatic transducers. Irrespective of this exists knowledge by epidemic studies about low-frequency electromagnetic fields and their health-impairing effects on biological organisms worldwide. A well-known result is the ‘‘TCO (’95) environmental labeling of displays’’ or radiation reduced computer monitors and screens. Independent efforts point out that the TCO idea should be valid also for other electronical consumer products like halogen headlights, clock radios, or housing machines. Is it necessary to add the near-field reinforcing headphone? First measurements over 23 headphones gave the answer, Yes [F. M. Koenig, 24. DAGA (1998), in preparation]. Using a pink noise C-weighted signal by a usual SPL of 70 dB reveals that the majority of customary headphones generate an electromagnetic field emission at the human head, which crosses the border level of the TCO’95 or German recommendation for the housing area [VDE-DIN No. 0848, part 4A1/11/90]. Continued investigations underline this earlier result. The steps’ contrary high-field dispositions will also be shown.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.