plemeri/InSPyReNet

Training Swin Transformer?

Closed this issue · 2 comments

I see that there are changes in comparison with the official implementation of Swin Transformer. Also is the Swin transformer getting trained while training Inspyrenet?

Hi, we follow the official implementation of Swin Transformer from their Semantic Segmentation version. Although we did not mention in the paper, there are some minor changes to provide intermediate feature maps.

We train both backbone while training our model, so if you use our method with swin transformer (InSPyReNet_SwinB), then every parameter in the network is participating to the training.

Thank you and feel free to ask more question 😄

Thank you.