By activating Google search, you confirm that you agree to the integration and the necessary data transmission to Google. If you want your browser to remember this decision, a cookie with a validity of 31 days will be stored.
Alternatively, you can also search with the free search engine DuckDuckGo!
Paper accepted for publication in the proceedings of BMVC2024
21.07.2024|14:18 Uhr
Our work with colleagues from EPFL's CVLab has been accepted for publication at BMVC, about which I am very glad! It is entitled "AttEntropy: Segmenting Unknown Objects in Complex Scenes using the Spatial Attention Entropy of Semantic Segmentation Transformers," authored by Krzysztof Lis, myself, Annika Mütze, Sina Honari, Pascal Fua & Mathieu Salzmann. We studied properties of attention-based segmentation transformers and had this cool finding that attention trained on semantic segmentation of street scenes can segment any object in any context. A soon to be updated version can be found here: https://arxiv.org/abs/2212.14397
