qubvel / segmentation_models

Segmentation models with pretrained backbones. Keras and TensorFlow Keras.
MIT License
4.73k stars 1.03k forks source link

Multi-label segmentation #302

Open sunwoo76 opened 4 years ago

sunwoo76 commented 4 years ago

I saw multi-class example in 'example dir', but I don't know this example will work well in the case of multi-label segmentation.

multi-label segmentation means that segmenting image that have pixels can have multi-label for example, [cancer, lesion, suspicious] = [1,0,1]. two true label is in one pixel.

which parts do I modify to train this case?

JordanMakesMaps commented 4 years ago
# multiclass segmentation with independent overlapping/non-overlapping class masks
model = sm.Unet('resnet34', classes=3, activation='sigmoid')

Maybe try this and see?