Using Pretrained Swin v2 model

vision
1

I tried following code to import swin v2 small :
image

I got following error:
image

I tried same approach for swin_s, which works out of the box. But when I am trying for swin_v2_s I get above error.

2

SwinV2 was adde ~2 months ago in this PR so you might need to update to a recent nightly binary in order to use it.

3

I installed nightly version using: pip3 install --pre torch torchvision torchaudio, which was listed here but still got the same error.
Could you please suggest any specific nightly binary version?

4

It works for me with the latest nightly:

>>> import torch
>>> import torchvision
>>> torch.__version__
'1.13.0.dev20221003+cu117'
>>> torchvision.__version__
'0.14.0.dev20221003+cu117'
>>> torchvision.models.swin_
torchvision.models.swin_b(           torchvision.models.swin_transformer  torchvision.models.swin_v2_t(
torchvision.models.swin_s(           torchvision.models.swin_v2_b(        
torchvision.models.swin_t(           torchvision.models.swin_v2_s(        
>>> torchvision.models.swin_v2_s
<function swin_v2_s at 0x7f3274c30160>
>>> model = torchvision.models.swin_v2_s()
>>> model
SwinTransformer(
  (features): Sequential(
    (0): Sequential(
      (0): Conv2d(3, 96, kernel_size=(4, 4), stride=(4, 4))
      (1): Permute()
      (2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
    )
    ...
5

Thank you very much @ptrblck . Now, its also working for me.