{"id":1996,"date":"2026-05-10T23:21:46","date_gmt":"2026-05-10T23:21:46","guid":{"rendered":"https:\/\/www.lalife.net\/?p=1996"},"modified":"2026-05-10T23:21:46","modified_gmt":"2026-05-10T23:21:46","slug":"u-net","status":"publish","type":"post","link":"https:\/\/www.lalife.net\/?p=1996","title":{"rendered":"U-Net"},"content":{"rendered":"\n

U-Net is a type of Convolutional Neural network with a U-shape structure. It’s used in analyzing medical image to find the outlier of target area. For example, it can be used to map out brain tumors.<\/p>\n\n\n\n

Why does it work? The first a few layers of convolutional network have more information of the edges, when the neural network going deeper, it finds more detailed information about the items in the image.<\/p>\n\n\n\n

Down1->Down2-> Down3-> Down4->Down5 -> Up5-> Up4-> Up3-> Up2-> Up1 <\/p>\n\n\n\n

Input : channel = 3<\/p>\n\n\n\n

Down1<\/strong>: 3×3 Conv2d, in_ch=3, out_ch= 64<\/td>Up1:<\/strong> concat down1 & Up2<\/td><\/tr>
3×3 Conv2d, in_ch=64, out_ch= 64<\/td>3×3 Conv2d in_ch=128 out_ch=64 + Relu<\/td><\/tr>
Relu activation after each Conv<\/td>3×3 Conv2d in_ch=64 out_ch=64 + Relu<\/td><\/tr>
2×2 Max Pooling ->1\/2 h, w<\/td>1×1 Conv2d in_ch=64 out_ch=64<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Down2<\/strong>: 3×3 Conv2d, in_ch=64 out_ch=128<\/td>Up2:<\/strong> concat down2 & Up3<\/td><\/tr>
3×3 Conv2d in_ch=64 out_ch=128<\/td>3×3 Conv2d in_ch=256 out_ch=128 + Relu<\/td><\/tr>
Relu activation after each Conv<\/td>3×3 Conv2d in_ch=128 out_ch=128 + Relu<\/td><\/tr>
2×2 Max Pooling<\/td>Upconv: filter kernel=2 stride=2<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Down3<\/strong>: 3×3 Conv2d in_ch=128 out_ch=256<\/td>Up3:<\/strong> concat down3 & Up4<\/td><\/tr>
3×3 Conv2d in_ch=256 out_ch=256<\/td>3×3 Conv2d in_ch=512 out_ch=256 + Relu<\/td><\/tr>
Relu activation after each Conv<\/td>3×3 Conv2d in_ch=256 out_ch=256 + Relu<\/td><\/tr>
2×2 Max Pooling <\/td>Upconv: filter kernel=2 stride=2<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Down4<\/strong>: 3×3 Conv2d in_ch=256 out_ch=512<\/td>Up4:<\/strong> concat down4 & Up 5<\/td><\/tr>
3×3 Conv2d in_ch=512 out_ch=512<\/td>3×3 Conv2d in_ch=1024 out_ch=512 + Relu<\/td><\/tr>
Relu activation after each Conv<\/td>3×3 Conv2d in_ch=512 out_ch=512+ Relu<\/td><\/tr>
2×2 Max Pooling <\/td>Upconv: filter kernel=2 stride=2<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Down<\/strong>5: 3×3 Conv2d in_ch=512 out_ch=1024<\/td>Up<\/strong>5: Upconv: filter kernel=2 stride=2<\/td><\/tr>
3×3 Conv2d in_ch=1024 out_ch= 024<\/td><\/td><\/tr>
Relu activation after each Conv<\/td><\/td><\/tr>
<\/td><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

References:<\/p>\n\n\n\n