How to input custom image size data, such as 64×64, when training my own dataset using the Swin-TransformerV2?

[Tismer]

When I use Swin-TransformerV2 to train my own dataset, the resolution of each image is 64×64 after data processing. However, it seems that I can't directly input them into the model as following error:

What should I do to ensure I can continue training? Could you please provide some suggestions?

[VyrtualL]

You have to change the img_size=224 (of ImageNet images,...) default value of the Swin-TransformerV2 you use. Because without changing that, the model think he get 224x224 images.

[UrGrandHappiness]

You have to change the img_size=224 (of ImageNet images,...) default value of the Swin-TransformerV2 you use. Because without changing that, the model think he get 224x224 images.

thank Q