Question about performance of smaller backbones

[QingCheng24]

Hi, thank you for the great work.

The results you reported are all based on Swin-L backbone and outperform Mask2Former.
Do the model with smaller backbones, e.g. ResNet50 and Swin-T, also outperform the former SOTA?
Or the text support requires the large backbone to show its improvement?

Best,