Summary of A Bayesian Approach to Ood Robustness in Image Classification, by Prakhar Kaushik and Adam Kortylewski and Alan Yuille
A Bayesian Approach to OOD Robustness in Image Classification
by Prakhar Kaushik, Adam Kortylewski, Alan Yuille
First submitted to arxiv on: 12 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Bayesian approach tackles the challenge of ensuring computer vision algorithms are robust to changes in image domains, particularly when no annotations are available from the target domain. Building upon Compositional Neural Networks (CompNets), which excel at occlusion but struggle with Out-of-Domain (OOD) data, the authors introduce Unsupervised Generative Transition (UGT). UGT leverages the generative head of CompNets, defined over feature vectors represented by von Mises-Fisher (vMF) kernels, to learn a transitional dictionary of vMF kernels intermediate between source and target domains. This approach is trained on source domain annotations and iteratively refined, demonstrating excellent performance in OOD scenarios, including occlusion, on various benchmarks such as the OOD-CV dataset, ImageNet-C, artificial image corruptions, and synthetic-to-real domain transfer. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper addresses a crucial problem in computer vision: making algorithms robust to changes in image domains. It’s like trying to recognize objects in a picture taken from a different angle or with different lighting. The solution uses a special type of neural network called Compositional Neural Networks, which are good at recognizing objects even when part of it is blocked. But what if we don’t have any examples of how the object looks in that new domain? The authors created a way to learn from the examples we do have and then apply it to the new domain. This approach works really well on different datasets and scenarios, including pictures with occlusion. |
Keywords
» Artificial intelligence » Neural network » Unsupervised
