search

Zian-Xu / Swin-MAE

Pytorch implementation of Swin MAE https://arxiv.org/abs/2212.13805
62 stars 12 forks source link

Why not use SwinV2 as Encoder? #1

Open ttjjmm opened 1 year ago

ttjjmm commented 1 year ago

SwinV2 which is an enhancing version based on SwinV1 has many merits beyond its predecesor, such as more stable training performance, using log pos_emb to support larger resolution, So I wonder why not use SwinV2 series structure for medical large-size image analysis?

Zian-Xu commented 1 year ago

This is a good suggestion. As a biomedical engineering master student, I am also learning all kinds of new methods. Perhaps replacing the backbone with a more novel network may give better results, which needs further experimental validation. Thank you for your question.