How do you calculate FLOPs?

[Kim-yhao]

Please tell me where is the code used to calculate FLOPs? Or can you tell me what is the shape of the input used to calculate FLOPs?

[Kim-yhao]

I tested my Swin-UMamba using thop.profile() and an image with shape (3, 224, 224), and the result was 4.4 GFLOPs, which is a big difference between your result.

[JiarunLiu]

Hi @Kim-yhao, we use fvcore to calculate FLOPs with the code from this issue. The shape of the input varies across different datasets:

• AbdomenMRI: (1, 320, 320)
• Endoscopy: (3, 384, 640)
• Microscopy: (3, 512, 512)

[Weixiang-Sun]

Hi @Kim-yhao, we use fvcore to calculate FLOPs with the code from this issue. The shape of the input varies across different datasets:

• AbdomenMRI: (1, 320, 320)
• Endoscopy: (3, 384, 640)
• Microscopy: (3, 512, 512)

Could you release the py file to calculate this in your project? thank you