Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation

Laura Yu Zheng, Wenjie Wei, Tony Wu, Jacob Clements, Shreelekha Revankar, Andre Harrison, Yu Shen, Ming Lin
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:78315-78337, 2025.

Abstract

Achieving robustness in image segmentation models is challenging due to the fine-grained nature of pixel-level classification. These models, which are crucial for many real-time perception applications, particularly struggle when faced with natural corruptions in the wild for autonomous systems. While sensitivity analysis can help us understand how input variables influence model outputs, its application to natural and uncontrollable corruptions in training data is computationally expensive. In this work, we present an adaptive, sensitivity-guided augmentation method to enhance robustness against natural corruptions. Our sensitivity analysis on average runs 10 times faster and requires about 200 times less storage than previous sensitivity analysis, enabling practical, on-the-fly estimation during training for a model-free augmentation policy. With minimal fine-tuning, our sensitivity-guided augmentation method achieves improved robustness on both real-world and synthetic datasets compared to state-of-the-art data augmentation techniques in image segmentation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zheng25n, title = {Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation}, author = {Zheng, Laura Yu and Wei, Wenjie and Wu, Tony and Clements, Jacob and Revankar, Shreelekha and Harrison, Andre and Shen, Yu and Lin, Ming}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {78315--78337}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zheng25n/zheng25n.pdf}, url = {https://proceedings.mlr.press/v267/zheng25n.html}, abstract = {Achieving robustness in image segmentation models is challenging due to the fine-grained nature of pixel-level classification. These models, which are crucial for many real-time perception applications, particularly struggle when faced with natural corruptions in the wild for autonomous systems. While sensitivity analysis can help us understand how input variables influence model outputs, its application to natural and uncontrollable corruptions in training data is computationally expensive. In this work, we present an adaptive, sensitivity-guided augmentation method to enhance robustness against natural corruptions. Our sensitivity analysis on average runs 10 times faster and requires about 200 times less storage than previous sensitivity analysis, enabling practical, on-the-fly estimation during training for a model-free augmentation policy. With minimal fine-tuning, our sensitivity-guided augmentation method achieves improved robustness on both real-world and synthetic datasets compared to state-of-the-art data augmentation techniques in image segmentation.} }
Endnote
%0 Conference Paper %T Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation %A Laura Yu Zheng %A Wenjie Wei %A Tony Wu %A Jacob Clements %A Shreelekha Revankar %A Andre Harrison %A Yu Shen %A Ming Lin %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zheng25n %I PMLR %P 78315--78337 %U https://proceedings.mlr.press/v267/zheng25n.html %V 267 %X Achieving robustness in image segmentation models is challenging due to the fine-grained nature of pixel-level classification. These models, which are crucial for many real-time perception applications, particularly struggle when faced with natural corruptions in the wild for autonomous systems. While sensitivity analysis can help us understand how input variables influence model outputs, its application to natural and uncontrollable corruptions in training data is computationally expensive. In this work, we present an adaptive, sensitivity-guided augmentation method to enhance robustness against natural corruptions. Our sensitivity analysis on average runs 10 times faster and requires about 200 times less storage than previous sensitivity analysis, enabling practical, on-the-fly estimation during training for a model-free augmentation policy. With minimal fine-tuning, our sensitivity-guided augmentation method achieves improved robustness on both real-world and synthetic datasets compared to state-of-the-art data augmentation techniques in image segmentation.
APA
Zheng, L.Y., Wei, W., Wu, T., Clements, J., Revankar, S., Harrison, A., Shen, Y. & Lin, M.. (2025). Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:78315-78337 Available from https://proceedings.mlr.press/v267/zheng25n.html.

Related Material