M.Sc. Simon Reiß
Vincenz-Priessnitz-Str. 3
Building 07.08
76131 Karlsruhe, Germany
About me
Hello there, my name is Simon, I work as a research assistant here at the Computer Vision for Human Computer Interaction Lab.
I am excited by democratizising machine learning and making it's benefits accessible to endeavours of all sizes.
In my research, I aim at breaking down the obstacles when deploying vision technology by designing cost-efficient, economical methods while upholding excellent performance.
I develop computer vision algorithms that leverage small and inexpensive data to solve semantic segmentation tasks.
If you as a passionate student are interested in bringing streamlined semantic segmentation systems to all developers regardless of how niche the field of application, shoot me an e-mail and come work with me (see below for open thesis topics).
Topics that interest me most include but are not limited to, semi-supervised learning, weakly-supervised learning, self-supervised learning, medical image segmentation as well as unsupervised image segmentation.
I work in a close collaboration with the Carl Zeiss AG.
Publications & Collaborations
Seibold, C.; Reiß, S.; Sarfraz, M. S.; Stiefelhagen, R.; Kleesiek, J.
2022. doi:10.5445/IR/1000146800
Seibold, C.; Reiß, S.; Kleesiek, J.; Stiefelhagen, R.
2022. Thirty-sixth AAAI conference on artificial intelligence. Online, 22.02.2022 - 01.03.2022, 2171–2179, Association for the Advancement of Artificial Intelligence (AAAI)
Reiß, S.; Roitberg, A.; Haurilet, M.; Stiefelhagen, R.
2020. 31st IEEE Intelligent Vehicles Symposium, IV 2020, Virtual, Las Vegas, United States, 19 October 2020 through 13 November 2020, 1042–1047, Institute of Electrical and Electronics Engineers (IEEE). doi:10.1109/IV47402.2020.9304782
Reiß, S.; Roitberg, A.; Haurilet, M.; Stiefelhagen, R.
2020. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2020; Virtual, Online; United States; 14 June 2020 through 19 June 2020, 3950–3955, Institute of Electrical and Electronics Engineers (IEEE). doi:10.1109/CVPRW50498.2020.00459
Martin, M.; Roitberg, A.; Haurilet, M.; Horne, M.; Reiß, S.; Voit, M.; Stiefelhagen, R.
2019. IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, Korea (South), 27 Oct.-2 Nov. 2019, 2801–2810, Institute of Electrical and Electronics Engineers (IEEE). doi:10.1109/ICCV.2019.00289
Open Thesis Topic: Semi-Supervised Semantic Segmentation via Robust Pseudo-labels and Reversible Image Transformations
- Subject:Semi-Supervised Semantic Segmentation
- Type:Master Thesis
- Tutor:
Simon Reiß
-
Checkout the full description of the topic here.
One component in semi-supervised segmentation methods can be inferring pseudo-labels for previously unlabeled (or weakly labeled) images. For a more robust pseudo-label generation, self-supervised literature might offer solutions. In this thesis, I would like to explore this exciting opportunity together with you.
Open Thesis Topic: Early Stopping without Pixel-wise Annotations for Semantic Segmentation with Scarce Data
- Subject:Semi-weakly Supervised Semantic Segmentation
- Type:Master Thesis
- Tutor:
Simon Reiß
-
Checkout the full description of the topic here.
Normal segmentation training procedures encompass using a training set and a validation set to find a good parameter configuration, i.e. selecting the right model checkpoint that performs best on the validation set. In practise having to use pixel-wise annotations for validation makes those costy labels inaccessible for training a better network. The question that poses itself is: Can we use more inexpensive annotations for validation or even unlabeled images?
Open Thesis Topic: Diverse Unsupervised Segmentation Proposals for Expert-driven Pixel-wise Annotation
- Subject:Unsupervised Segmentation
- Type:Master Thesis
- Tutor:
Simon Reiß
-
Checkout the full description of the topic here.
Unsupervised semantic segmentation is a very challenging task, yet it offers the possiblity to pre-compute regions within an image that might be helpful in the annotation process. An expert could select these regions instead of having to annotate the images pixel-wise. To achieve this properly, it would be advantageous if multiple sensible regions can be proposed by the same algorithm. Can we design an unsupervised segmentation model that achives this?