我要的是这样的%% Horizontal UNet Architecture Diagram
%% Generated from Python code with features=[64, 128, 256]
flowchart LR
subgraph Encoder
input[Input\nHxWx1] --> enc1_conv
subgraph enc1["Encoder Block 1"]
enc1_conv[Conv3x3, ReLU\n64 channels] --> enc1_conv2[Conv3x3, ReLU\n64 channels]
end
enc1_conv2 --> pool1[MaxPool2x2]:::pool
enc1_conv2 --> skip1[Skip Connection 1]:::skip
pool1 --> enc2_conv
subgraph enc2["Encoder Block 2"]
enc2_conv[Conv3x3, ReLU\n128 channels] --> enc2_conv2[Conv3x3, ReLU\n128 channels]
end
enc2_conv2 --> pool2[MaxPool2x2]:::pool
enc2_conv2 --> skip2[Skip Connection 2]:::skip
pool2 --> enc3_conv
subgraph enc3["Encoder Block 3"]
enc3_conv[Conv3x3, ReLU\n256 channels] --> enc3_conv2[Conv3x3, ReLU\n256 channels]
end
enc3_conv2 --> pool3[MaxPool2x2]:::pool
enc3_conv2 --> skip3[Skip Connection 3]:::skip
end
subgraph Bottleneck
pool3 --> bottle_conv
subgraph bottle["Bottleneck"]
bottle_conv[Conv3x3, ReLU\n512 channels] --> bottle_conv2[Conv3x3, ReLU\n512 channels]
end
end
subgraph Decoder
bottle_conv2 --> dec1_up[Transposed Conv2x2\n256 channels]:::upconv
skip3 --> dec1_cat[Concatenation]:::concat
dec1_up --> dec1_cat
dec1_cat --> dec1_conv
subgraph dec1["Decoder Block 1"]
dec1_conv[Conv3x3, ReLU\n256 channels] --> dec1_conv2[Conv3x3, ReLU\n256 channels]
end
dec1_conv2 --> dec2_up[Transposed Conv2x2\n128 channels]:::upconv
skip2 --> dec2_cat[Concatenation]:::concat
dec2_up --> dec2_cat
dec2_cat --> dec2_conv
subgraph dec2["Decoder Block 2"]
dec2_conv[Conv3x3, ReLU\n128 channels] --> dec2_conv2[Conv3x3, ReLU\n128 channels]
end
dec2_conv2 --> dec3_up[Transposed Conv2x2\n64 channels]:::upconv
skip1 --> dec3_cat[Concatenation]:::concat
dec3_up --> dec3_cat
dec3_cat --> dec3_conv
subgraph dec3["Decoder Block 3"]
dec3_conv[Conv3x3, ReLU\n64 channels] --> dec3_conv2[Conv3x3, ReLU\n64 channels]
end
end
dec3_conv2 --> final_conv[1x1 Conv\nOutput Channels]:::final
final_conv --> softmax[Softmax]:::softmax
softmax --> output[Output\nHxWxN]
%% Styling
classDef conv fill:#9f9,stroke:#333,stroke-width:2px;
classDef pool fill:#f99,stroke:#333,stroke-width:2px;
classDef skip fill:#ff9,stroke:#333,stroke-width:2px,stroke-dasharray:5 5;
classDef upconv fill:#99f,stroke:#333,stroke-width:2px;
classDef concat fill:#f9f,stroke:#333,stroke-width:2px;
classDef final fill:#f66,stroke:#333,stroke-width:2px;
classDef softmax fill:#6f6,stroke:#333,stroke-width:2px;
class enc1_conv,enc1_conv2,enc2_conv,enc2_conv2,enc3_conv,enc3_conv2,bottle_conv,bottle_conv2,dec1_conv,dec1_conv2,dec2_conv,dec2_conv2,dec3_conv,dec3_conv2 conv;
class pool1,pool2,pool3 pool;
class skip1,skip2,skip3 skip;
class dec1_up,dec2_up,dec3_up upconv;
class dec1_cat,dec2_cat,dec3_cat concat;
class final_conv final;
class softmax softmax;
%% Connections
linkStyle 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27 stroke:#333,stroke-width:2px;
最新发布
本文介绍了解决TensorFlow-gpu使用过程中遇到的内存不足问题。通过排查cudnn版本兼容性和释放内存资源的方法，成功解决了运行时出现的错误。
扫一扫
1915
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
