Paper:RemoteNet: Remote Sensing Image Segmentation Network based on Global-Local Information
Because the original paper does not have open source code, I tried to reproduce it.
😒😒😒 Although the paper does not say it, the Encoder is PVT_v2,so the code was modified!!!😢😢
torch==2.1.1+cu118
torchaudio==2.1.1+cu118
torchinfo==1.8.0
torchvision==0.16.1
requests==2.28.1
urllib3== 1.25.11
timm==0.9.0 #It must be ensured that timm is this version!!!python train.py
Modify the path before this
train_dataset = PotsdamDataset(data_root='potsdam', mode='train',img_dir='images/train', mask_dir='anns/train',
mosaic_ratio=0.25, transform=train_aug)
val_dataset = PotsdamDataset(data_root='potsdam',img_dir='images/val', mask_dir='anns/val',transform=val_aug)https://www.123pan.com/s/YnBgjv-D8j1H.html提取码:pots
https://drive.google.com/file/d/1NWLLVlUVaIZYwePbPx7Ca60AdyW-cdMT/view?usp=sharing
==============================================================================================================
Layer (type:depth-idx) Output Shape Param #
==============================================================================================================
RemoteNet [1, 6, 1024, 1024] --
├─Encoder: 1-1 [1, 64, 256, 256] --
│ └─FeatureListNet: 2-1 [1, 64, 256, 256] --
│ │ └─OverlapPatchEmbed: 3-1 [1, 256, 256, 64] 9,600
│ │ └─PyramidVisionTransformerStage: 3-2 [1, 64, 256, 256] 701,056
│ │ └─PyramidVisionTransformerStage: 3-3 [1, 128, 128, 128] 1,279,616
│ │ └─PyramidVisionTransformerStage: 3-4 [1, 320, 64, 64] 3,682,880
│ │ └─PyramidVisionTransformerStage: 3-5 [1, 512, 32, 32] 7,822,848
├─Decoder: 1-2 [1, 6, 1024, 1024] --
│ └─Conv2d: 2-2 [1, 64, 32, 32] 32,832
│ └─Block: 2-3 [1, 64, 32, 32] --
│ │ └─BatchNorm2d: 3-6 [1, 64, 32, 32] 128
│ │ └─GlobalLocalAttention: 3-7 [1, 64, 32, 32] 63,816
│ │ └─Identity: 3-8 [1, 64, 32, 32] --
│ │ └─BatchNorm2d: 3-9 [1, 64, 32, 32] 128
│ │ └─Mlp: 3-10 [1, 64, 32, 32] 33,088
│ │ └─Identity: 3-11 [1, 64, 32, 32] --
│ │ └─ConvBN: 3-12 [1, 64, 32, 32] 4,224
│ │ └─ConvBN: 3-13 [1, 64, 32, 32] 36,992
│ │ └─ConvBN: 3-14 [1, 64, 32, 32] 102,528
│ │ └─SeparableConvBN: 3-15 [1, 64, 32, 32] 7,360
│ └─Fusion: 2-4 [1, 64, 64, 64] --
│ │ └─Conv2d: 3-16 [1, 64, 64, 64] 20,544
│ │ └─Sequential: 3-17 [1, 64, 1, 1] 41,344
│ │ └─Sequential: 3-18 [1, 64, 1, 1] (recursive)
│ │ └─ConvBNReLU: 3-19 [1, 64, 64, 64] 36,992
│ │ └─ConvBNReLU: 3-20 [1, 64, 64, 64] (recursive)
│ └─Block: 2-5 [1, 64, 64, 64] --
│ │ └─BatchNorm2d: 3-21 [1, 64, 64, 64] 128
│ │ └─GlobalLocalAttention: 3-22 [1, 64, 64, 64] 63,816
│ │ └─Identity: 3-23 [1, 64, 64, 64] --
│ │ └─BatchNorm2d: 3-24 [1, 64, 64, 64] 128
│ │ └─Mlp: 3-25 [1, 64, 64, 64] 33,088
│ │ └─Identity: 3-26 [1, 64, 64, 64] --
│ │ └─ConvBN: 3-27 [1, 64, 64, 64] 4,224
│ │ └─ConvBN: 3-28 [1, 64, 64, 64] 36,992
│ │ └─ConvBN: 3-29 [1, 64, 64, 64] 102,528
│ │ └─SeparableConvBN: 3-30 [1, 64, 64, 64] 7,360
│ └─Fusion: 2-6 [1, 64, 128, 128] --
│ │ └─Conv2d: 3-31 [1, 64, 128, 128] 8,256
│ │ └─Sequential: 3-32 [1, 64, 1, 1] 41,344
│ │ └─Sequential: 3-33 [1, 64, 1, 1] (recursive)
│ │ └─ConvBNReLU: 3-34 [1, 64, 128, 128] 36,992
│ │ └─ConvBNReLU: 3-35 [1, 64, 128, 128] (recursive)
│ └─Block: 2-7 [1, 64, 128, 128] --
│ │ └─BatchNorm2d: 3-36 [1, 64, 128, 128] 128
│ │ └─GlobalLocalAttention: 3-37 [1, 64, 128, 128] 63,816
│ │ └─Identity: 3-38 [1, 64, 128, 128] --
│ │ └─BatchNorm2d: 3-39 [1, 64, 128, 128] 128
│ │ └─Mlp: 3-40 [1, 64, 128, 128] 33,088
│ │ └─Identity: 3-41 [1, 64, 128, 128] --
│ │ └─ConvBN: 3-42 [1, 64, 128, 128] 4,224
│ │ └─ConvBN: 3-43 [1, 64, 128, 128] 36,992
│ │ └─ConvBN: 3-44 [1, 64, 128, 128] 102,528
│ │ └─SeparableConvBN: 3-45 [1, 64, 128, 128] 7,360
│ └─Fusion: 2-8 [1, 64, 256, 256] --
│ │ └─Conv2d: 3-46 [1, 64, 256, 256] 4,160
│ │ └─Sequential: 3-47 [1, 64, 1, 1] 41,344
│ │ └─Sequential: 3-48 [1, 64, 1, 1] (recursive)
│ │ └─ConvBNReLU: 3-49 [1, 64, 256, 256] 36,992
│ │ └─ConvBNReLU: 3-50 [1, 64, 256, 256] (recursive)
│ └─FRM: 2-9 [1, 6, 1024, 1024] --
│ │ └─SeparableConvBN: 3-51 [1, 64, 256, 256] 4,800
│ │ └─SeparableConvBN: 3-52 [1, 64, 256, 256] 5,824
│ │ └─ConvBN: 3-53 [1, 64, 256, 256] 4,224
│ │ └─ConvBNReLU: 3-54 [1, 64, 256, 256] 36,992
│ │ └─SeparableConvBNReLU: 3-55 [1, 64, 256, 256] 103,104
│ │ └─ConvBN: 3-56 [1, 64, 256, 256] (recursive)
│ │ └─ConvBNReLU: 3-57 [1, 64, 256, 256] 4,224
│ │ └─ConvBNReLU: 3-58 [1, 64, 256, 256] (recursive)
│ │ └─Conv2d: 3-59 [1, 6, 256, 256] 390
==============================================================================================================
Total params: 14,701,150
Trainable params: 14,701,150
Non-trainable params: 0
Total mult-adds (G): 34.69
==============================================================================================================
Input size (MB): 12.58
Forward/backward pass size (MB): 4018.96
Params size (MB): 58.78
Estimated Total Size (MB): 4090.32
==============================================================================================================