方法对比

	图像分类	Vision Transformer
领域	深度学习	深度学习
方法族	Machine learning	Machine learning
起源年份	2012 (deep CNN era); conceptual roots 1989 (LeCun)	2021
提出者	Krizhevsky, A.; Sutskever, I.; Hinton, G. E.	Dosovitskiy, A. et al.
类型	Supervised classification task	Transformer architecture for images (self-attention over patches)
开创性文献	Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems (NeurIPS), 25, 1097–1105.	Dosovitskiy, A. et al. (2021). An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. ICLR.
别名	visual classification, image recognition, CNN-based classification, visual categorization	Görsel Transformer (ViT), görsel transformer, ViT, patch transformer for images
相关	5	5
摘要	Image classification is the task of assigning a single semantic label to an entire image from a fixed set of categories. Modern approaches rely on deep convolutional neural networks (CNNs) or Vision Transformers (ViTs) trained end-to-end on large labeled datasets such as ImageNet, achieving superhuman accuracy on many benchmarks and underpinning applications from medical imaging to autonomous vehicles.	The Vision Transformer (ViT), introduced by Dosovitskiy and colleagues in 2021, splits an image into fixed-size patches, treats those patches as a sequence, and applies the Transformer self-attention mechanism to image classification. Given enough training data, it surpasses convolutional neural networks (CNNs).