Burf / SwinTransformer-Tensorflow2

SwinTransformer for Tensorflow2

SwinTransformer for Tensorflow2

This is an implementation of "SwinTransformer V1 and V2" on Keras and Tensorflow.

The implementation is based on papers[1, 2] and official implementation[3].

Model

• Model
  • SwinTransformer-Tiny
  • SwinTransformer-Small
  • SwinTransformer-Base
  • SwinTransformer-Large
  • SwinTransformerV2-Tiny
  • SwinTransformerV2-Small
  • SwinTransformerV2-Base
  • SwinTransformerV2-Large
• Pre-trained weight(The pre-trained weights are converted from official weight.)
  • imagenet 1k (Tiny, Small, Base)
  • imagenet 22k (Base, Large)
  • imagenet 22kto1k (Base, Large)

Requirements

• Python 3
• tensorflow 2
• torch 1.1▲ (Use the pre-trained weights)

Reference

1. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows, Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo, https://arxiv.org/abs/2103.14030v2

2. Swin Transformer V2: Scaling Up Capacity and Resolution, Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo, https://arxiv.org/abs/2111.09883

3. Swin Transformer, microsoft, https://github.com/microsoft/Swin-Transformer

Contributor

• Hyungjin Kim([e-mail])