segmodel.py

import segmentation_models as sm
from UNetPlusPlus.segmentation_models import Xnet

from keras.optimizers import *
from utils import jaccard_distance_loss

# define model

def denseunet(pretrained_weights=None, lr=1E-4):
    model = sm.Unet('densenet121', encoder_weights='imagenet')

    model.compile(
        optimizer=Adam(lr),
        loss=sm.losses.bce_jaccard_loss,
        metrics=[sm.metrics.iou_score],
    )
    if pretrained_weights:
        model.load_weights(pretrained_weights)
    return model


def unetxx(pretrained_weights=None, lr=1E-4):
    model = Xnet(backbone_name='resnet50', encoder_weights=None, decoder_block_type='transpose')
    model.compile(optimizer=Adam(lr=lr),
                  loss=jaccard_distance_loss,
                  metrics=[sm.metrics.iou_score, 'accuracy'])
    if pretrained_weights:
        model.load_weights(pretrained_weights)
    return model