search

rishizek / tensorflow-deeplab-v3-plus

DeepLabv3+ built in TensorFlow
MIT License
834 stars 307 forks source link

Binary Image Semantic Segmentation #37

Open ghost opened 5 years ago

ghost commented 5 years ago

I have a problem of image segmentation in which I have only one class means its mask and other is background. I have changed the number of classes in model and to 2 but it show miou=1 and result are worst.

bonbonjour commented 5 years ago

I want to train my own data with 16 class,where should be changed in the code ? Thanks

margokhokhlova commented 5 years ago

@Murad1997 Did you solve this? I am curently working on the same issue, I think, the softmax somewhere should be replaced by a sigmoid for a binary classification problem, but I am not sure

LbinB commented 5 years ago

did you solve the problem? i am face the same issue,could you help me if you already know how to do @rishizek @margokhokhlova @bonbonjour @Murad1997

margokhokhlova commented 5 years ago

@LbinB I used this code for binary segmentation, fine-tuning the thing, but I didn't get very good results (a bit less accurate than a simple unet). The number of classes even for a binary segmentation was 2, and I just the replaced last activation layer.