Swin Transformer Code

In this tutorial, I skimmed through the paper ‘Swin Transformer: Hierarchical Vision Transformer using Shifted Windows.’ I also run the code on a Jupiter notebook with some printouts that show how the tensor changes inside the transformer.
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
🔗 swin transformer annotated: https://github.com/mashaan14/YouTube-...
🔗 swin transformer training: https://github.com/mashaan14/YouTube-...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
- website https://mashaan14.github.io/mashaan/
- github https://github.com/mashaan14
- X https://x.com/mashaan_14
- linkedin   / mashaan  
- sponsor me on GitHub Sponsors https://github.com/sponsors/mashaan14
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
📹 Video edit: Adobe Premiere Rush
🎧 Audio enhancement: Adobe Podcast
🖼️ Thumbnails: GIMP
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Chapters:
00:00 start
01:10 the hierarchical structure in image classification
05:00 swin transformer github repository
05:38 transformer architecture
06:53 start of the code
08:35 running a single batch into the transformer
12:49 patch merging
17:40 2nd, 3rd, and 4th layers of the transformer
18:57 a note about patch merging
23:31 layer norm
30:27 an example on layer norm
31:42 swin transformer blocks
33:50 shifted windows code
35:43 attention mask
37:45 model summary
40:29 are you interested in a jax version of this?
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
#Swin #SwinTransformer #ViT #attention #multiheadattention #convolution #ai #deeplearning #machinelearning #python #neuralnetworks #artificialintelligence #pytorch #torchvision #convolutionalneuralnetworks #transformers #image #imageclassification #computervision #tutorial