Anderes
Permanent URI for this collection
Browse
Recent Submissions
1 - 5 of 1192
-
PublicationFine-Grained Product Classification on Leaflet Advertisements( 2023)In this paper, we describe a first publicly available fine-grained product recognition dataset based on leaflet images. Using advertisement leaflets, collected over several years from different European retailers, we provide a total of 41.6k manually annotated product images in 832 classes. Further, we investigate three different approaches for this fine-grained product classification task, Classification by Image, by Text, as well as by Image and Text. The approach "Classification by Text" uses the text extracted directly from the leaflet product images. We show, that the combination of image and text as input improves the classification of visual difficult to distinguish products. The final model leads to an accuracy of 96.4% with a Top-3 score of 99.2%.
-
PublicationAn Extended Study of Human-like Behavior under Adversarial Training( 2023)Neural networks have a number of shortcomings. Amongst the severest ones is the sensitivity to distribution shifts which allows models to be easily fooled into wrong predictions by small perturbations to inputs that are often imperceivable to humans and do not have to carry semantic meaning. Adversarial training poses a partial solution to address this issue by training models on worst-case perturbations. Yet, recent work has also pointed out that the reasoning in neural networks is different from humans. Humans identify objects by shape, while neural nets mainly employ texture cues. Exemplarily, a model trained on photographs will likely fail to generalize to datasets containing sketches. Interestingly, it was also shown that adversarial training seems to favorably increase the shift toward shape bias. In this work, we revisit this observation and provide an extensive analysis of this effect on various architectures. Further, we provide a possible explanation for this phenomenon from a frequency perspective.
-
Publication„Begehbares Internet“ - die Vortragsreihe „Fokus Zukunft - Unser Leben 2050“ diskutiert Chancen und Risiken des Metaverse( 2022)Wie überdimensionale Tauchermasken sehen sie aus: und tatsächlich kann man mit den Headsets der Virtual Reality in digitale Parallelwelten eintauchen. Dagegen erlauben Datenbrillen wie Google Glass, digitale Zusatzinformationen in die Wahrnehmung der analogen Wirklichkeit einzublenden. Beides - Augmented Reality und Virtual Reality - sind Varianten des sogenannten Metaverse. Mit ihm wird das Internet dreidimensional und sozusagen begehbar. Im Fall der Virtual Reality schicken wir in der Form eines Avatars sogar einen digitalen Stellvertreter von uns selbst in den Cyberspace. Das klingt futuristisch, aber es wird intensiv daran gearbeitet. Denn ein begehbares Internet bietet ungeahnte Möglichkeiten für Ausbildung, Marketing und Social Media.
-
Publication„Standortvorteil durch Quantencomputing“ - die Vortragsreihe „Fokus: Zukunft. Unser Leben 2050.“ thematisiert die zweite Quantenrevolution( 2022)Vor 122 Jahren entdeckte Max Planck, dass in der unendlich kleinen Welt der Atome ganz andere Gesetze gelten als sonst in der Physik. In der Folge bescherte uns die sogenannte Quantenmechanik so revolutionäre Erfindungen wie den Laser oder die Satellitennavigation. Für das 21. Jahrhundert bereitet sich jetzt eine zweite Quantenrevolution vor. Die Physiker haben inzwischen gelernt, auch einzelne Atome hin- und herzuschieben und damit Quantencomputer zu bauen. Damit werden sich in Zukunft hochkomplizierte Aufgaben rechnen lassen, für die heutige Supercomputer noch viele, viele Jahren benötigen würden.
-
PublicationGetting To Tape-Out Quicker With Analog Layout Generators( 2019)Generators for schematics, test benches, simulation control, and layouts can significantly increase efficiency in the various design phases. All design engineers know it well: there is hardly any time left until tape-out, but the amount of work that remains is not decreasing as fast as the deadline is approaching. The intricate schematic must still be implemented as a layout, and many recurring tasks slow down the progress. The real crux often lies in specific parts of the circuit - parts that often have lower performance demands but are necessary and take up a great deal of time in the design stage. Can automation help to save time here?
