On the inconsistency between the printed network structure and the paper

[PowerKaly]

Hello, author
I added at the end of models/fasternet.py

if __name__ == "__main__":
    model = FasterNet(
        mlp_ratio=2.0,
        embed_dim=128,
        depths=(1, 2, 13, 2),
        drop_path_rate=0.15,
        act_layer='RELU',
        fork_feat=True,
    )

    print(model)

Try to print the network structure, and the result is

If for detection, please install mmdetection first
FasterNet(
  (patch_embed): PatchEmbed(
    (proj): Conv2d(3, 128, kernel_size=(4, 4), stride=(4, 4), bias=False)
    (norm): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  )
  (stages): Sequential(
    (0): BasicStage(
      (blocks): Sequential(
        (0): MLPBlock(
          (drop_path): Identity()
          (mlp): Sequential(
            (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
            (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
            (2): ReLU(inplace=True)
            (3): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
          )
          (spatial_mixing): Partial_conv3(
            (partial_conv3): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          )
        )
      )
    )
    (1): PatchMerging(
      (reduction): Conv2d(128, 256, kernel_size=(2, 2), stride=(2, 2), bias=False)
      (norm): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    )
    (2): BasicStage(
      (blocks): Sequential(
        (0): MLPBlock(
          (drop_path): DropPath(drop_prob=0.009)
          (mlp): Sequential(
            (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
            (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
            (2): ReLU(inplace=True)
            (3): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          )
          (spatial_mixing): Partial_conv3(
            (partial_conv3): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          )
        )
        (1): MLPBlock(
          (drop_path): DropPath(drop_prob=0.018)
          (mlp): Sequential(
            (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
            (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
            (2): ReLU(inplace=True)
            (3): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          )
          (spatial_mixing): Partial_conv3(
            (partial_conv3): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          )
        )
      )
    )
    (3): PatchMerging(
      (reduction): Conv2d(256, 512, kernel_size=(2, 2), stride=(2, 2), bias=False)
      (norm): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    )
........

As you can see from here.

(mlp): Sequential(
           (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
           (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
           (2): ReLU(inplace=True)
           (3): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
         )
         (spatial_mixing): Partial_conv3(
           (partial_conv3): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
         )

It goes through two 1x1 convolutions and then 3x3. Contrary to the paper, I don't know if there is something wrong with my understanding. I hope you can give me some advice.thinks 👍 :)