remove useless code

Author: zsdonghao

main.py

@@ -1,7 +1,6 @@
 #! /usr/bin/python
 # -*- coding: utf8 -*-
 
-
 import os, time, pickle, random, time
 from datetime import datetime
 import numpy as np
@@ -14,9 +13,6 @@
 from utils import *
 from config import config, log_config
 
-
-
-
 ###====================== HYPER-PARAMETERS ===========================###
 ## Adam
 batch_size = config.TRAIN.batch_size
@@ -87,8 +83,6 @@ def train():
 
     net_vgg, vgg_target_emb = Vgg19_simple_api((t_target_image_224+1)/2, reuse=False)
     _, vgg_predict_emb = Vgg19_simple_api((t_predict_image_224+1)/2, reuse=True)
-    # print(vgg_predict_emb.outputs)
-    # # exit()
 
     ## test inference
     net_g_test = SRGAN_g(t_image, is_train=False, reuse=True)
@@ -98,31 +92,11 @@ def train():
     d_loss2 = tl.cost.sigmoid_cross_entropy(logits_fake, tf.zeros_like(logits_fake), name='d2')
     d_loss = d_loss1 + d_loss2
 
-    # g_gan_loss = 1e-1 * tl.cost.sigmoid_cross_entropy(logits_fake, tf.ones_like(logits_fake), name='g')
-    # mse_loss = normalize_mean_squared_error(net_g.outputs, t_target_image)                                            # simiao
-    # vgg_loss = 5e-1 * normalize_mean_squared_error(vgg_predict_emb.outputs, vgg_target_emb.outputs)
-
-    g_gan_loss = 1e-3 * tl.cost.sigmoid_cross_entropy(logits_fake, tf.ones_like(logits_fake), name='g')                 # paper 1e-3
-    mse_loss = tl.cost.mean_squared_error(net_g.outputs , t_target_image, is_mean=True)                                 # paper
-    vgg_loss = 2e-6 * tl.cost.mean_squared_error(vgg_predict_emb.outputs, vgg_target_emb.outputs, is_mean=True)    # simiao
-
-    ## simiao
-    # g_gan_loss = tl.cost.sigmoid_cross_entropy(logits_fake, tf.ones_like(logits_fake), name='g')
-    # mse_loss = normalize_mean_squared_error(net_g.outputs, t_target_image)
-    # vgg_loss = 0.00025 * tl.cost.mean_squared_error(vgg_predict_emb.outputs, vgg_target_emb.outputs, is_mean=True)
-
-    ## history
-    # resize-conv MSE + 1e-2*g_gan_loss: 1020 green broken, but can recover/ 1030 always green
-    # resize-conv MSE + 1e-3*g_gan_loss: more stable than 1e-2, 1043 bubble
-    # resize-conv MSE + 1e-3*g_gan_loss +1e-6*VGG 相比 mse+gan, bubble少了很多，d loss ≈ 0.5 (G not powerful?)
-    # subpixel-conv MSE + 1e-3*g_gan_loss +1e-6*VGG (no pretrain), small checkboard. VGG loss ≈ MSE / 2
-    # train higher VGG loss?
-    # subpixel-conv MSE + 1e-3*g_gan_loss +2e-6*VGG (no pretrain), small checkboard. VGG loss ≈ MSE
-    # subpixel-conv MSE + 1e-4*g_gan_loss +2e-6*VGG (no pretrain), small checkboard. 50epoch d loss very small ≈ 0.02054373
-    # subpixel-conv MSE + 1e-3*g_gan_loss +2e-6*VGG, 100 epoch pretrain, bare checkboard!
+    g_gan_loss = 1e-3 * tl.cost.sigmoid_cross_entropy(logits_fake, tf.ones_like(logits_fake), name='g')
+    mse_loss = tl.cost.mean_squared_error(net_g.outputs , t_target_image, is_mean=True)
+    vgg_loss = 2e-6 * tl.cost.mean_squared_error(vgg_predict_emb.outputs, vgg_target_emb.outputs, is_mean=True)
 
     g_loss = mse_loss + vgg_loss + g_gan_loss
-    # g_loss = mse_loss + g_gan_loss
 
     g_vars = tl.layers.get_variables_with_name('SRGAN_g', True, True)
     d_vars = tl.layers.get_variables_with_name('SRGAN_d', True, True)
@@ -305,6 +279,7 @@ def evaluate():
 
     size = valid_lr_img.shape
     t_image = tf.placeholder('float32', [None, size[0], size[1], size[2]], name='input_image')
+    # t_image = tf.placeholder('float32', [1, None, None, 3], name='input_image')
 
     net_g = SRGAN_g(t_image, is_train=False, reuse=False)