Swin transformer, the best paper model of iccv 2021, finally started with video!

publisher ：Towhee Technical team

In the last year Swin Transformer get ICCV 2021 After the best paper , Microsoft Research Asia launched another masterpiece in the video field this year Video Swin Transformer .**Video Swin Transformer Model in CVPR 2022 Neutral can kill the list , It is superior to ViViT、TimeSformer Wait for the Internet ！** The model is in Kinetics-400 The accuracy on is 84.9% top-1, stay Something-Something v2 The accuracy on is 69.6% top-1.

Video Swin Transformer: an illustrated example of 3D shifted windows

Expand from image field to video field ,Swin Transformer Introduce local inductive bias on the basis of the original , And effectively use the pre training image model . Compared with before ,Video Swin Transformer Even spatiotemporal decomposition can globally calculate self attention , Therefore, we can better balance speed and accuracy .

Reference material ：

Model use case ：action-classification/video-swin-transformer

Address of thesis ：Video Swin Transformer

Official statement ： Tu Bang video understands several major tasks ！ Microsoft put forward ：Video Swin Transformer

Other information ： Video classification weapon Video Swin Transformer

For more project updates and details, please pay attention to our project ( https://github.com/towhee-io/towhee/blob/main/towhee/models/README_CN.md) , Your attention is a powerful driving force for us to generate electricity with love , welcome star, fork, slack Three even :)