How to train Unet semantic segmentation with only one single class/label?

13 次查看（过去 30 天）
Al
Al 2024-8-6
评论： Matt J 2024-8-6
Hello, I'm currently working on a task to do a semantic segmentation on USG image to locate TMJ. I did my image labelling in Image Labeler App and only did one class so that the class region will be 1 and background is 0. I was about to train my model with unetLayers but it says "The value of 'numClasses' is invalid. Expected numClasses to be a scalar with value > 1."
I'm aware that someone asked this similar question here with an answer, but I want to ask, how to customize specifically unetLayers to accomodate single class? I see that unetLayers also has softmaxLayers but i can't find the pixel layers. Thank you in advance!

请先登录，再回答此问题。

回答（1 个）

Matt J
Matt J 2024-8-6
编辑：Matt J 2024-8-6
Ran in:
The value of 'numClasses' is invalid. Expected numClasses to be a scalar with value > 1
Because you have two classes (0 and 1). There is no such thing as a single class.
I see that unetLayers also has softmaxLayers but i can't find the pixel layers. Thank you in advance!
Where are you looking? The final layer should be the Pixel Classification Layer, as in the following example,
imageSize = [480 640 3];
numClasses = 5;
encoderDepth = 3;
lgraph = unetLayers(imageSize,numClasses,'EncoderDepth',encoderDepth);
lgraph.Layers(end-5:end)
ans =
6x1 Layer array with layers: 1 'Decoder-Stage-3-ReLU-1' ReLU ReLU 2 'Decoder-Stage-3-Conv-2' 2-D Convolution 64 3x3 convolutions with stride [1 1] and padding 'same' 3 'Decoder-Stage-3-ReLU-2' ReLU ReLU 4 'Final-ConvolutionLayer' 2-D Convolution 5 1x1 convolutions with stride [1 1] and padding 'same' 5 'Softmax-Layer' Softmax softmax 6 'Segmentation-Layer' Pixel Classification Layer Cross-entropy loss
  2 个评论
Al
Al 2024-8-6
编辑：Al 2024-8-6
Thank you so much! I think I was panicking with my assignment so I didn't properly notice the layers
Matt J
Matt J 2024-8-6
You're welcome, but please Accept-click the answer to indicate that it resolved your quesiton.

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by