Computer Science
5 years ago

Transformers outperform CNNs in image recognition, requiring less computational power, revolutionizing computer vision.

arXiv (Cornell University)
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby

Paper Summary

Related papers

Theme background image
Engineering
4 years ago

New method enables robots to pick up objects with low contrast images.

Autofocusing
Curve Detection
Stereo Vision
2 views
Theme background image
Engineering
5 years ago

New method revolutionizes robot grasping of transparent objects with precision.

SLAM
Fabric Detection
Curve Detection
1 view
Theme background image
Computer Science
7 years ago

Revolutionizing robot vision: 3D object recognition made simple with video segmentation.

Stereo Vision
Autofocusing
Image Correlation
1 view

Transformers, known for language tasks, can also work well for image recognition without needing convolutional networks. By using a pure transformer on image patches, the Vision Transformer (ViT) achieves great results on various image classification tests like ImageNet and CIFAR-100. ViT needs less computing power compared to traditional convolutional networks.