Backbone output shape

[liketheflower]

The original images has the of height 530 and width of 730.

>>> img.shape
(530, 730, 3)
>>> 800/530
1.509433962264151
>>> 730*(800/530)
1101.8867924528304

The input image is converted to

input shape torch.Size([1, 3, 800, 1120])

Resnet-50

>>> for i in range(4):
...     print(df["outs"][i].shape)
... 
torch.Size([1, 256, 200, 280])
torch.Size([1, 512, 100, 140])
torch.Size([1, 1024, 50, 70])
torch.Size([1, 2048, 25, 35])

Swin_T

>>> df = pd.read_pickle("swin_outs.pkl")
>>> for i in range(4):
...     print(df["outs"][i].shape)
... 
torch.Size([1, 96, 200, 280])
torch.Size([1, 192, 100, 140])
torch.Size([1, 384, 50, 70])
torch.Size([1, 768, 25, 35])