Input images of other sizes

[Zhcyoung]

I found that all input images require transforms.Resize(224). Can CHIEF extract features without resizing images?Currently, when I input images with other size, an error occurs.

model = ctranspath()
model.head = nn.Identity()
td = torch.load(r'./model_weights/CHIEF_CTransPath.pth')
model.load_state_dict(td['model'], strict=True)
batch = torch.randn((1,3,512,512))
features = model(batch)

AssertionError: Input image size (512*512) doesn't match model (224*224).

[Xiyue-Wang]

Hi, Zhcyoung, yes. To use the pre-trained weights, the input patches need to be resized to 224x224 pixels.

[Zhcyoung]

ok,Thank you.