Skip to content

Changes

Page history
Update JULAIN Journal Club authored by susanne wenzel's avatar susanne wenzel
Hide whitespace changes
Inline Side-by-side
JULAIN-Journal-Club.md
View page @ 6a0e8fc8
...@@ -56,10 +56,12 @@ Nikita Araslanov, Computer Vision Group at TU Munich in Germany ...@@ -56,10 +56,12 @@ Nikita Araslanov, Computer Vision Group at TU Munich in Germany
The high accuracy of modern semantic segmentation models hinges on expensive high-quality dense annotation. Therefore, it is of high practical importance to design unsupervised objectives to learn semantic representations that generalise well to real data. This talk will focus on one principle towards this goal: semantic equivariance. The underlying idea is to exploit equivariance of the semantic maps to similarity transformations w.r.t. the input image. We will consider specific implementations and extensions of this technique in two specific problem formulations in computer vision. First, we will take a look at the unsupervised domain adaptation, where we adapt our model, trained on annotated synthetic data, to unlabelled real-world images. In the second example leveraging the equivariance, I will present a simple approach that substantially improves the segmentation accuracy on out-of-distribution data. In this setting, there is no target distribution available for model adaptation as before, but only a single datum from that distribution. I will highlight some of the limitations and conclude the presentation with an outlook on promising directions. The high accuracy of modern semantic segmentation models hinges on expensive high-quality dense annotation. Therefore, it is of high practical importance to design unsupervised objectives to learn semantic representations that generalise well to real data. This talk will focus on one principle towards this goal: semantic equivariance. The underlying idea is to exploit equivariance of the semantic maps to similarity transformations w.r.t. the input image. We will consider specific implementations and extensions of this technique in two specific problem formulations in computer vision. First, we will take a look at the unsupervised domain adaptation, where we adapt our model, trained on annotated synthetic data, to unlabelled real-world images. In the second example leveraging the equivariance, I will present a simple approach that substantially improves the segmentation accuracy on out-of-distribution data. In this setting, there is no target distribution available for model adaptation as before, but only a single datum from that distribution. I will highlight some of the limitations and conclude the presentation with an outlook on promising directions.
#### Papers #### Papers
* [[Self-supervised augmentation consistency for adapting semantic segmentation](https://scholar.google.com/citations?view_op=view_citation&hl=en&user=RdMFioAAAAAJ&sortby=pubdate&citation_for_view=RdMFioAAAAAJ:WF5omc3nYNoC)] * [Self-supervised augmentation consistency for adapting semantic segmentation](https://scholar.google.com/citations?view_op=view_citation&hl=en&user=RdMFioAAAAAJ&sortby=pubdate&citation_for_view=RdMFioAAAAAJ:WF5omc3nYNoC)<br>
Nikita Araslanov, Stefan Roth, CVPR 2021 Nikita Araslanov, Stefan Roth, CVPR 2021
* [[Semantic Self-adaptation: Enhancing Generalization with a Single Sample ](https://scholar.google.com/citations?view_op=view_citation&hl=en&user=RdMFioAAAAAJ&citation_for_view=RdMFioAAAAAJ:LkGwnXOMwfcC)] * [Semantic Self-adaptation: Enhancing Generalization with a Single Sample ](https://scholar.google.com/citations?view_op=view_citation&hl=en&user=RdMFioAAAAAJ&citation_for_view=RdMFioAAAAAJ:LkGwnXOMwfcC)<br>
* Invitation and moderation: Hanno Scharr arXiv preprint
**Invitation and moderation**: Hanno Scharr
### 17 October 2022 ### 17 October 2022
......