keras-perceptual_loss

最新推荐文章于 2023-08-20 12:24:31 发布

googler_offer

最新推荐文章于 2023-08-20 12:24:31 发布

阅读量962

点赞数

分类专栏：深度学习=====keras搬砖系列

本文链接：https://blog.csdn.net/googler_offer/article/details/82913927

版权

深度学习=====keras搬砖系列专栏收录该内容

55 篇文章 5 订阅

订阅专栏

keras-perceptual_loss

L1, L2 Loss

def l1_loss(y_true,y_pred):
    return K.mean(K.abs(y_true-y_pred))

def l2_loss(y_true,y_pred):
    return K.mean(K.square(y_true-y_pred))

Perceptual Loss

def perceptual_loss(y_true, y_pred):  # y_true and y_pred's pixels are scaled between 0 to 255
    y_true = preprocess_input(y_true)
    y_pred = preprocess_input(y_pred)
    vgg = VGG19(include_top=False, weights='imagenet', input_shape=(256,256,3))
    loss_model = Model(inputs=vgg.input, outputs=vgg.get_layer('block3_conv3').output)
    loss_model.trainable = False
    return K.mean(K.square(loss_model(y_true)-loss_model(y_pred)))