Self-supervised pretrained weights

[fleecedragoon]

Hello,

I'm interested in making use of a self-supervised pretrained model on CT imaging. In the tutorials, I found two options: ViT_UNETR_SSL or Pre-trained SwinUNETR Backbone.

I have two questions regarding these options:

1. Are the pretrained weights for the ViT_UNETR_SSL available without having to download the COVID dataset and pretraining the modified ViT?

2. In the SwinUNETR tutorial, it seems like it should be possible to download the weights, but the path to the where the weights should be stored does not exist

use_pretrained = True
if use_pretrained:
    resource = (
        "https://github.com/Project-MONAI/MONAI-extra-test-data/releases/download/0.8.1/ssl_pretrained_weights.pth"
    )
    dst = "./ssl_pretrained_weights.pth"
    download_url(resource, dst)
    pretrained_path = os.path.normpath(dst)

[KumoLiu]

Hi @fleecedragoon, the "ssl_pretrained_weights.pth" is located here. Thanks!

[fleecedragoon]

Thanks!

[fleecedragoon]

Hi again,

What is the difference between the model_swinvit.pt weights, which should correspond to "the self-supervised pre-trained weights for Swin UNETR backbone" and the ssl_pretrained_weights.pth weights, which also correspond to the pre-trained weights for SwinUNETR?