Certified Training for Convolutional Perturbations
2026-07-20 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionMachine Learning
AI summaryⓘ
The authors studied how vision models can fail when images get blurry, like from a shaky camera, which is a problem for important uses like object detection. They point out that usual fixes like data augmentation or adversarial training don't provide guaranteed safety. So, they created a new training method that can prove the model is robust against certain blurs. Their approach works better than past methods, achieving high accuracy even when images are reasonably blurry, without losing regular accuracy.
vision modelsmotion blurdata augmentationadversarial trainingrobustnesscertified trainingconvolutional perturbationsCIFAR10object detectionrobust accuracy
Authors
Benedikt Brückner, Alessio Lomuscio
Abstract
Vision models have been found to be susceptible to perturbations such as motion blur induced at runtime by a shaking camera. This impedes their deployment in critical applications since phenomena such as slightly blurred vision might lead to failures, for example an object detector missing objects. While methods such as data augmentation or Adversarial Training can improve empirical robustness, they lack formal safety guarantees, making it difficult to identify and mitigate hidden vulnerabilities. We introduce a novel Certified Training approach that leverages an efficient encoding of convolutional perturbations to train provably robust models. Our method significantly outperforms Adversarial Training, achieving, for example, over 80% robust accuracy against motion blur of reasonable intensity on CIFAR10 while maintaining comparable standard accuracy.