大数据开发之keras代码框架应用

     总体来讲keras这个深度学习框架真的很“简易”，它体现在可参考的文档写的比较详细，不像caffe，装完以后都得靠技术博客，keras有它自己的官方文档（不过是英文的），这给初学者提供了很大的学习空间。

    在此做下代码框架应用笔记

     

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70

class VGGNetwork:     def append_vgg_network(self, x_in, true_X_input):         return x #x is output of VGG     def load_vgg_weight(self, model):         return model class DiscriminatorNetwork:     def append_gan_network(self, true_X_input):         return x class GenerativeNetwork:     def create_sr_model(self, ip):         return x     def get_generator_output(self, input_img, srgan_model):         return self.output_func([input_img]) class SRGANNetwork:     def build_srgan_pretrain_model(self):         return self.srgan_model_     def build_discriminator_pretrain_model(self):         return self.discriminative_model_     def build_srgan_model(self):         return self.srgan_model_     def pre_train_srgan(self, image_dir, nb_images=50000, nb_epochs=1, use_small_srgan=False):         for i in range(nb_epochs):             for x in datagen.flow_from_directory                 if iteration % 50 == 0 and iteration != 0                     validation//print psnr                 Train only generator + vgg network                 if iteration % 1000 == 0 and iteration != 0                     Saving model weights     def pre_train_discriminator(self, image_dir, nb_images=50000, nb_epochs=1, batch_size=128):         for i in range(nb_epochs):              for x in datagen.flow_from_directory                  Train only discriminator                  if iteration % 1000 == 0 and iteration != 0                     Saving model weights     def train_full_model(self, image_dir, nb_images=50000, nb_epochs=10):           for i in range(nb_epochs):             for x in datagen.flow_from_directory                 if iteration % 50 == 0 and iteration != 0                     validation//print psnr                 if iteration % 1000 == 0 and iteration != 0                     Saving model weights                 Train only discriminator, disable training of srgan                 Train only generator, disable training of discriminator if __name__ == "__main__":     from keras.utils.visualize_util import plot       # Path to MS COCO dataset     coco_path = r"D:\Yue\Documents\Dataset\coco2014\train2014"       '''     Base Network manager for the SRGAN model       Width / Height = 32 to reduce the memory requirement for the discriminator.       Batch size = 1 is slower, but uses the least amount of gpu memory, and also acts as     Instance Normalization (batch norm with 1 input image) which speeds up training slightly.     '''       srgan_network = SRGANNetwork(img_width=32, img_height=32, batch_size=1)     srgan_network.build_srgan_model()     #plot(srgan_network.srgan_model_, 'SRGAN.png', show_shapes=True)       # Pretrain the SRGAN network     #srgan_network.pre_train_srgan(coco_path, nb_images=80000, nb_epochs=1)       # Pretrain the discriminator network     #srgan_network.pre_train_discriminator(coco_path, nb_images=40000, nb_epochs=1, batch_size=16)       # Fully train the SRGAN with VGG loss and Discriminator loss     srgan_network.train_full_model(coco_path, nb_images=80000, nb_epochs=5)