B
Björn Barz
Researcher at University of Jena
Publications - 38
Citations - 507
Björn Barz is an academic researcher from University of Jena. The author has contributed to research in topics: Image retrieval & Content-based image retrieval. The author has an hindex of 8, co-authored 38 publications receiving 291 citations. Previous affiliations of Björn Barz include Schiller International University.
Papers
More filters
Proceedings ArticleDOI
Deep Learning on Small Datasets without Pre-Training using Cosine Loss
Björn Barz,Joachim Denzler +1 more
TL;DR: It is shown that the cosine loss function provides substantially better performance than crossentropy on datasets with only a handful of samples per class, and integrating prior knowledge in the form of class hierarchies is straightforward with thecosine loss and improves classification performance further.
Journal ArticleDOI
Detecting Regions of Maximal Divergence for Spatio-Temporal Anomaly Detection
TL;DR: The “Maximally Divergent Intervals” (MDI) framework for unsupervised detection of coherent spatial regions and time intervals characterized by a high Kullback-Leibler divergence compared with all other data given is proposed.
Journal ArticleDOI
Do We Train on Test Data? Purging CIFAR of Near-Duplicates
Björn Barz,Joachim Denzler +1 more
TL;DR: The "fair CIFAR" (ciFAIR) dataset is provided, where all duplicates in the test sets are replaced with new images sampled from the same domain, and whether recent research has overfitted to memorizing data instead of learning abstract concepts is investigated.
Proceedings ArticleDOI
Hierarchy-Based Image Embeddings for Semantic Image Retrieval
Björn Barz,Joachim Denzler +1 more
TL;DR: This paper proposed to map images onto class embeddings whose pairwise dot products correspond to a measure of semantic similarity between classes, which not only improve image retrieval results, but could also facilitate integrating semantics for other tasks, e.g., novelty detection or few-shot learning.
Proceedings ArticleDOI
Hierarchy-based Image Embeddings for Semantic Image Retrieval.
Björn Barz,Joachim Denzler +1 more
TL;DR: This work introduces a deterministic algorithm for computing the class centroids directly based on prior world-knowledge encoded in a hierarchy of classes such as WordNet, and shows that learned semantic image embeddings improve the semantic consistency of image retrieval results by a large margin.