$S^{2}$-FracMix enhances deep visual models with novel augmentation strategy

Researchers have introduced $S^{2}$-FracMix, a novel data augmentation technique designed to improve the generalization of deep visual models. This method constructs challenging yet label-consistent samples by extracting salient image patches and reinserting them into different regions of the same image, promoting scale-invariant feature learning without cross-sample interference. The framework further enhances robustness by incorporating self-similarity patterns, enabling simultaneous learning from both fractal and non-fractal structures within an image. Empirical evaluations across seven benchmarks demonstrate that $S^{2}$-FracMix achieves state-of-the-art performance in various tasks, including classification, robustness, and object detection. AI

IMPACT This new augmentation strategy could lead to more robust and accurate deep visual models across various applications.

RANK_REASON The item is a research paper detailing a new data augmentation technique for computer vision models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• Connected Papers
• CORE Recommender
• DagsHub
• FracMix
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• $S^{2}$-FracMix
• ScienceCast
• scite Smart Citations
• Self-Saliency Mixup

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →