19 Sep 2024
SUMix, developed by researchers primarily from Chongqing Technology and Business University, introduces a 'plug-and-play' framework that enhances cutting-based Mixup data augmentation techniques. This method dynamically learns a semantic-aware mix ratio and models uncertainty, consistently improving classification accuracy and robustness against occlusion and corruptions across various datasets and deep learning architectures.
There are no more papers matching your filters at the moment.
