init

README.md

@@ -1,19 +1,58 @@
-#### 从命令行创建一个新的仓库
+# KinD
+This is a Tensorflow implementation of KinD
 
-```bash
-touch README.md
-git init
-git add README.md
-git commit -m "first commit"
-git remote add origin https://bdgit.educoder.net/ZhengHui/KinD.git
-git push -u origin master
+### The [KinD++](https://github.com/zhangyhuaee/KinD_plus) is an improved version. ###
 
-```
+Kindling the Darkness: a Practical Low-light Image Enhancer. In ACMMM2019<br>
+Yonghua Zhang, Jiawan Zhang, Xiaojie Guo
+
+### [Paper](http://doi.acm.org/10.1145/3343031.3350926)
+<img src="figures/network.jpg" width="800px"/>
+
+<img src="figures/result.jpg" width="800px"/>
+
+### Requirements ###
+1. Python
+2. Tensorflow >= 1.10.0
+3. numpy, PIL
 
-#### 从命令行推送已经创建的仓库
+### Test ###
+First download the pre-trained checkpoints from [BaiduNetdisk](https://pan.baidu.com/s/1c4ZLYEIoR-8skNMiAVbl_A) or [google drive](https://drive.google.com/open?id=1-ljWntl7FExf6BSQtl5Mz3rMGWgnXDz4), then just run
+```shell
+python evaluate.py
+```
+Our pre-trained model has changed. Thus, the results have some difference with the report in our paper. However, you can adjust the illumination ratio to get better results.
 
-```bash
-git remote add origin https://bdgit.educoder.net/ZhengHui/KinD.git
-git push -u origin master
+### Train ###
+The original LOLdataset can be downloaded from [here](https://daooshee.github.io/BMVC2018website/). We rearrange the original LOLdataset and add several all-zero images to improve the decomposition results and restoration results. The new dataset can be download from [BaiduNetdisk](https://pan.baidu.com/s/1sn3vWJ2I5U2dlVUD7eqIBQ) or [google drive](https://drive.google.com/open?id=1-MaOVG7ylOkmGv1K4HWWcrai01i_FeDK). Save training pairs of LOL dataset under './LOLdataset/our485/' and save evaluating pairs under './LOLdataset/eval15/'. For training, just run
+```shell
+python decomposition_net_train.py
+python adjustment_net_train.py
+python reflectance_restoration_net_train.py
+```
+You can also evaluate on the LOLdataset, just run
+```shell
+python evaluate_LOLdataset.py
+```
+Our code partly refers to the [code](https://github.com/weichen582/RetinexNet).
 
+### Citation ###
+```
+@inproceedings{zhang2019kindling,
+ author = {Zhang, Yonghua and Zhang, Jiawan and Guo, Xiaojie},
+ title = {Kindling the Darkness: A Practical Low-light Image Enhancer},
+ booktitle = {Proceedings of the 27th ACM International Conference on Multimedia},
+ series = {MM '19},
+ year = {2019},
+ isbn = {978-1-4503-6889-6},
+ location = {Nice, France},
+ pages = {1632--1640},
+ numpages = {9},
+ url = {http://doi.acm.org/10.1145/3343031.3350926},
+ doi = {10.1145/3343031.3350926},
+ acmid = {3350926},
+ publisher = {ACM},
+ address = {New York, NY, USA},
+ keywords = {image decomposition, image restoration, low light enhancement},
+}
 ```

adjustment_net_train.py

@@ -0,0 +1,223 @@
+# coding: utf-8
+from __future__ import print_function
+import os
+import time
+import random
+#from skimage import color
+from PIL import Image
+import tensorflow as tf
+import numpy as np
+from utils import *
+from model import *
+from glob import glob
+
+batch_size = 10
+patch_size = 48
+
+sess = tf.Session()
+#the input of decomposition net
+input_decom = tf.placeholder(tf.float32, [None, None, None, 3], name='input_decom')
+#the input of illumination adjustment net
+input_low_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i')
+input_low_i_ratio = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i_ratio')
+input_high_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_high_i')
+
+[R_decom, I_decom] = DecomNet_simple(input_decom)
+#the output of decomposition network
+decom_output_R = R_decom
+decom_output_I = I_decom
+#the output of illumination adjustment net
+output_i = Illumination_adjust_net(input_low_i, input_low_i_ratio)
+
+#define loss
+
+def grad_loss(input_i_low, input_i_high):
+    x_loss = tf.square(gradient(input_i_low, 'x') - gradient(input_i_high, 'x'))
+    y_loss = tf.square(gradient(input_i_low, 'y') - gradient(input_i_high, 'y'))
+    grad_loss_all = tf.reduce_mean(x_loss + y_loss)
+    return grad_loss_all
+
+loss_grad = grad_loss(output_i, input_high_i)
+loss_square = tf.reduce_mean(tf.square(output_i  - input_high_i))# * ( 1 - input_low_r ))#* (1- input_low_i)))
+
+loss_adjust =  loss_square + loss_grad 
+
+lr = tf.placeholder(tf.float32, name='learning_rate')
+
+optimizer = tf.train.AdamOptimizer(learning_rate=lr, name='AdamOptimizer')
+
+var_Decom = [var for var in tf.trainable_variables() if 'DecomNet' in var.name]
+var_adjust = [var for var in tf.trainable_variables() if 'Illumination_adjust_net' in var.name]
+
+saver_adjust = tf.train.Saver(var_list=var_adjust)
+saver_Decom = tf.train.Saver(var_list = var_Decom)
+train_op_adjust = optimizer.minimize(loss_adjust, var_list = var_adjust)
+sess.run(tf.global_variables_initializer())
+print("[*] Initialize model successfully...")
+
+### load data
+### Based on the decomposition net, we first get the decomposed reflectance maps 
+### and illumination maps, then train the adjust net.
+###train_data
+train_low_data = []
+train_high_data = []
+train_low_data_names = glob('./LOLdataset/our485/low/*.png') 
+train_low_data_names.sort()
+train_high_data_names = glob('./LOLdataset/our485/high/*.png') 
+train_high_data_names.sort()
+assert len(train_low_data_names) == len(train_high_data_names)
+print('[*] Number of training data: %d' % len(train_low_data_names))
+for idx in range(len(train_low_data_names)):
+    low_im = load_images(train_low_data_names[idx])
+    train_low_data.append(low_im)
+    high_im = load_images(train_high_data_names[idx])
+    train_high_data.append(high_im)
+
+pre_decom_checkpoint_dir = './checkpoint/decom_net_train/'
+ckpt_pre=tf.train.get_checkpoint_state(pre_decom_checkpoint_dir)
+if ckpt_pre:
+    print('loaded '+ckpt_pre.model_checkpoint_path)
+    saver_Decom.restore(sess,ckpt_pre.model_checkpoint_path)
+else:
+    print('No pre_decom_net checkpoint!')
+
+#decomposed_low_r_data_480 = []
+decomposed_low_i_data_480 = []
+#decomposed_high_r_data_480 = []
+decomposed_high_i_data_480 = []
+for idx in range(len(train_low_data)):
+    input_low = np.expand_dims(train_low_data[idx], axis=0)
+    RR, II = sess.run([decom_output_R, decom_output_I], feed_dict={input_decom: input_low})
+    RR0 = np.squeeze(RR)
+    II0 = np.squeeze(II)
+    print(RR0.shape, II0.shape)
+    #decomposed_high_r_data_480.append(result_1_sq)
+    decomposed_low_i_data_480.append(II0)
+for idx in range(len(train_high_data)):
+    input_high = np.expand_dims(train_high_data[idx], axis=0)
+    RR2, II2 = sess.run([decom_output_R, decom_output_I], feed_dict={input_decom: input_high})
+    RR02 = np.squeeze(RR2)
+    II02 = np.squeeze(II2)
+    print(RR02.shape, II02.shape)
+    #decomposed_high_r_data_480.append(result_1_sq)
+    decomposed_high_i_data_480.append(II02)
+
+eval_adjust_low_i_data = decomposed_low_i_data_480[451:480]
+eval_adjust_high_i_data = decomposed_high_i_data_480[451:480]
+
+train_adjust_low_i_data = decomposed_low_i_data_480[0:450]
+train_adjust_high_i_data = decomposed_high_i_data_480[0:450]
+
+print('[*] Number of training data: %d' % len(train_adjust_high_i_data))
+
+learning_rate = 0.0001
+epoch = 2000
+eval_every_epoch = 200
+train_phase = 'adjustment'
+numBatch = len(train_adjust_low_i_data) // int(batch_size)
+train_op = train_op_adjust
+train_loss = loss_adjust
+saver = saver_adjust
+
+checkpoint_dir = './checkpoint/illumination_adjust_net_train/'
+if not os.path.isdir(checkpoint_dir):
+    os.makedirs(checkpoint_dir)
+ckpt=tf.train.get_checkpoint_state(checkpoint_dir)
+if ckpt:
+    print('loaded '+ckpt.model_checkpoint_path)
+    saver.restore(sess,ckpt.model_checkpoint_path)
+else:
+    print("No adjustment net pre model!")
+
+start_step = 0
+start_epoch = 0
+iter_num = 0
+print("[*] Start training for phase %s, with start epoch %d start iter %d : " % (train_phase, start_epoch, iter_num))
+
+sample_dir = './illumination_adjust_net_train/'
+if not os.path.isdir(sample_dir):
+    os.makedirs(sample_dir)
+
+start_time = time.time()
+image_id = 0
+
+for epoch in range(start_epoch, epoch):
+    for batch_id in range(start_step, numBatch):
+        batch_input_low_i_ratio = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        batch_input_high_i_ratio = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        batch_input_low_i = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        batch_input_high_i = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        input_low_i_rand = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        input_high_i_rand = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        input_low_i_rand_ratio = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+        input_high_i_rand_ratio = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")
+
+        for patch_id in range(batch_size):
+            i_low_data = train_adjust_low_i_data[image_id]
+            i_low_expand = np.expand_dims(i_low_data, axis = 2)
+            i_high_data = train_adjust_high_i_data[image_id]
+            i_high_expand = np.expand_dims(i_high_data, axis = 2)
+
+            h, w = train_adjust_low_i_data[image_id].shape
+            x = random.randint(0, h - patch_size)
+            y = random.randint(0, w - patch_size)
+            i_low_data_crop = i_low_expand[x : x+patch_size, y : y+patch_size, :]
+            i_high_data_crop = i_high_expand[x : x+patch_size, y : y+patch_size, :]
+
+            rand_mode = np.random.randint(0, 7)
+            batch_input_low_i[patch_id, :, :, :] = data_augmentation(i_low_data_crop , rand_mode)
+            batch_input_high_i[patch_id, :, :, :] = data_augmentation(i_high_data_crop, rand_mode)
+
+            ratio = np.mean(i_low_data_crop/(i_high_data_crop+0.0001))
+            #print(ratio)
+            i_low_data_ratio = np.ones([patch_size,patch_size])*(1/ratio+0.0001)
+            i_low_ratio_expand = np.expand_dims(i_low_data_ratio , axis =2)
+            i_high_data_ratio = np.ones([patch_size,patch_size])*(ratio)
+            i_high_ratio_expand = np.expand_dims(i_high_data_ratio , axis =2)
+            batch_input_low_i_ratio[patch_id, :, :, :] = i_low_ratio_expand
+            batch_input_high_i_ratio[patch_id, :, :, :] = i_high_ratio_expand
+
+            rand_mode = np.random.randint(0, 2)
+            if rand_mode == 1:
+                input_low_i_rand[patch_id, :, :, :] = batch_input_low_i[patch_id, :, :, :]
+                input_high_i_rand[patch_id, :, :, :] = batch_input_high_i[patch_id, :, :, :]
+                input_low_i_rand_ratio[patch_id, :, :, :] = batch_input_low_i_ratio[patch_id, :, :, :]
+                input_high_i_rand_ratio[patch_id, :, :, :] = batch_input_high_i_ratio[patch_id, :, :, :]
+            else:
+                input_low_i_rand[patch_id, :, :, :] = batch_input_high_i[patch_id, :, :, :]
+                input_high_i_rand[patch_id, :, :, :] = batch_input_low_i[patch_id, :, :, :]
+                input_low_i_rand_ratio[patch_id, :, :, :] = batch_input_high_i_ratio[patch_id, :, :, :]
+                input_high_i_rand_ratio[patch_id, :, :, :] = batch_input_low_i_ratio[patch_id, :, :, :]
+
+            image_id = (image_id + 1) % len(train_adjust_low_i_data)
+
+        _, loss = sess.run([train_op, train_loss], feed_dict={input_low_i: input_low_i_rand,input_low_i_ratio: input_low_i_rand_ratio,\
+                                                              input_high_i: input_high_i_rand, \
+                                                              lr: learning_rate})
+        print("%s Epoch: [%2d] [%4d/%4d] time: %4.4f, loss: %.6f" \
+              % (train_phase, epoch + 1, batch_id + 1, numBatch, time.time() - start_time, loss))
+        iter_num += 1
+    if (epoch + 1) % eval_every_epoch == 0:
+        print("[*] Evaluating for phase %s / epoch %d..." % (train_phase, epoch + 1))
+        
+        for idx in range(10):
+            rand_idx = idx#np.random.randint(26)
+            input_uu_i = eval_adjust_low_i_data[rand_idx] 
+            input_low_eval_i = np.expand_dims(input_uu_i, axis=0)
+            input_low_eval_ii = np.expand_dims(input_low_eval_i, axis=3)
+            h, w = eval_adjust_low_i_data[idx].shape
+            rand_ratio = np.random.random(1)*2
+            input_uu_i_ratio = np.ones([h,w]) * rand_ratio 
+            input_low_eval_i_ratio = np.expand_dims(input_uu_i_ratio, axis=0)
+            input_low_eval_ii_ratio = np.expand_dims(input_low_eval_i_ratio, axis=3)
+            
+            result_1 = sess.run(output_i, feed_dict={input_low_i: input_low_eval_ii, input_low_i_ratio: input_low_eval_ii_ratio})
+            save_images(os.path.join(sample_dir, 'h_eval_%d_%d_%5f.png' % ( epoch + 1 , rand_idx + 1,   rand_ratio)), input_uu_i, result_1)
+        
+
+    saver.save(sess, checkpoint_dir + 'model.ckpt')
+
+print("[*] Finish training for phase %s." % train_phase)
+
+
+

checkpoint/Restoration_net_train/checkpoint

@@ -0,0 +1 @@
+model_checkpoint_path: "model.ckpt"

checkpoint/decom_net_train/checkpoint

@@ -0,0 +1,2 @@
+model_checkpoint_path: "model.ckpt"
+all_model_checkpoint_paths: "model.ckpt"

checkpoint/illumination_adjust_net_train/checkpoint

@@ -0,0 +1,2 @@
+model_checkpoint_path: "model.ckpt"
+all_model_checkpoint_paths: "model.ckpt"

decomposition_net_train.py

@@ -0,0 +1,174 @@
+# coding: utf-8
+from __future__ import print_function
+import os, time, random
+import tensorflow as tf
+from PIL import Image
+import numpy as np
+from utils import *
+from model import *
+from glob import glob
+
+batch_size = 10
+patch_size = 48
+
+sess = tf.Session()
+
+input_low = tf.placeholder(tf.float32, [None, None, None, 3], name='input_low')
+input_high = tf.placeholder(tf.float32, [None, None, None, 3], name='input_high')
+
+[R_low, I_low] = DecomNet_simple(input_low)
+[R_high, I_high] = DecomNet_simple(input_high)
+
+I_low_3 = tf.concat([I_low, I_low, I_low], axis=3)
+I_high_3 = tf.concat([I_high, I_high, I_high], axis=3)
+
+#network output
+output_R_low = R_low
+output_R_high = R_high
+output_I_low = I_low_3
+output_I_high = I_high_3
+
+# define loss
+
+def mutual_i_loss(input_I_low, input_I_high):
+    low_gradient_x = gradient(input_I_low, "x")
+    high_gradient_x = gradient(input_I_high, "x")
+    x_loss = (low_gradient_x + high_gradient_x)* tf.exp(-10*(low_gradient_x+high_gradient_x))
+    low_gradient_y = gradient(input_I_low, "y")
+    high_gradient_y = gradient(input_I_high, "y")
+    y_loss = (low_gradient_y + high_gradient_y) * tf.exp(-10*(low_gradient_y+high_gradient_y))
+    mutual_loss = tf.reduce_mean( x_loss + y_loss) 
+    return mutual_loss
+
+def mutual_i_input_loss(input_I_low, input_im):
+    input_gray = tf.image.rgb_to_grayscale(input_im)
+    low_gradient_x = gradient(input_I_low, "x")
+    input_gradient_x = gradient(input_gray, "x")
+    x_loss = tf.abs(tf.div(low_gradient_x, tf.maximum(input_gradient_x, 0.01)))
+    low_gradient_y = gradient(input_I_low, "y")
+    input_gradient_y = gradient(input_gray, "y")
+    y_loss = tf.abs(tf.div(low_gradient_y, tf.maximum(input_gradient_y, 0.01)))
+    mut_loss = tf.reduce_mean(x_loss + y_loss) 
+    return mut_loss
+
+recon_loss_low = tf.reduce_mean(tf.abs(R_low * I_low_3 -  input_low))
+recon_loss_high = tf.reduce_mean(tf.abs(R_high * I_high_3 - input_high))
+
+equal_R_loss = tf.reduce_mean(tf.abs(R_low - R_high))
+
+i_mutual_loss = mutual_i_loss(I_low, I_high)
+
+i_input_mutual_loss_high = mutual_i_input_loss(I_high, input_high)
+i_input_mutual_loss_low = mutual_i_input_loss(I_low, input_low)
+
+loss_Decom = 1*recon_loss_high + 1*recon_loss_low \
+               + 0.01 * equal_R_loss + 0.2*i_mutual_loss \
+             + 0.15* i_input_mutual_loss_high + 0.15* i_input_mutual_loss_low
+
+###
+lr = tf.placeholder(tf.float32, name='learning_rate')
+
+optimizer = tf.train.AdamOptimizer(learning_rate=lr, name='AdamOptimizer')
+var_Decom = [var for var in tf.trainable_variables() if 'DecomNet' in var.name]
+
+train_op_Decom = optimizer.minimize(loss_Decom, var_list = var_Decom)
+sess.run(tf.global_variables_initializer())
+
+saver_Decom = tf.train.Saver(var_list = var_Decom)
+print("[*] Initialize model successfully...")
+
+#load data
+###train_data
+train_low_data = []
+train_high_data = []
+train_low_data_names = glob('./LOLdataset/our485/low/*.png') 
+train_low_data_names.sort()
+train_high_data_names = glob('./LOLdataset/our485/high/*.png') 
+train_high_data_names.sort()
+assert len(train_low_data_names) == len(train_high_data_names)
+print('[*] Number of training data: %d' % len(train_low_data_names))
+for idx in range(len(train_low_data_names)):
+    low_im = load_images(train_low_data_names[idx])
+    train_low_data.append(low_im)
+    high_im = load_images(train_high_data_names[idx])
+    train_high_data.append(high_im)
+###eval_data
+eval_low_data = []
+eval_high_data = []
+eval_low_data_name = glob('./LOLdataset/eval15/low/*.png')
+eval_low_data_name.sort()
+eval_high_data_name = glob('./LOLdataset/eval15/high/*.png*')
+eval_high_data_name.sort()
+for idx in range(len(eval_low_data_name)):
+    eval_low_im = load_images(eval_low_data_name[idx])
+    eval_low_data.append(eval_low_im)
+    eval_high_im = load_images(eval_high_data_name[idx])
+    eval_high_data.append(eval_high_im)
+
+
+epoch = 2000
+learning_rate = 0.0001
+
+sample_dir = './Decom_net_train/'
+if not os.path.isdir(sample_dir):
+    os.makedirs(sample_dir)
+
+eval_every_epoch = 200
+train_phase = 'decomposition'
+numBatch = len(train_low_data) // int(batch_size)
+train_op = train_op_Decom
+train_loss = loss_Decom
+saver = saver_Decom
+
+checkpoint_dir = './checkpoint/decom_net_train/'
+if not os.path.isdir(checkpoint_dir):
+    os.makedirs(checkpoint_dir)
+ckpt=tf.train.get_checkpoint_state(checkpoint_dir)
+if ckpt:
+    print('loaded '+ckpt.model_checkpoint_path)
+    saver.restore(sess,ckpt.model_checkpoint_path)
+
+start_step = 0
+start_epoch = 0
+iter_num = 0
+print("[*] Start training for phase %s, with start epoch %d start iter %d : " % (train_phase, start_epoch, iter_num))
+
+start_time = time.time()
+image_id = 0
+for epoch in range(start_epoch, epoch):
+    for batch_id in range(start_step, numBatch):
+        batch_input_low = np.zeros((batch_size, patch_size, patch_size, 3), dtype="float32")
+        batch_input_high = np.zeros((batch_size, patch_size, patch_size, 3), dtype="float32")
+        for patch_id in range(batch_size):
+            h, w, _ = train_low_data[image_id].shape
+            x = random.randint(0, h - patch_size)
+            y = random.randint(0, w - patch_size)
+            rand_mode = random.randint(0, 7)
+            batch_input_low[patch_id, :, :, :] = data_augmentation(train_low_data[image_id][x : x+patch_size, y : y+patch_size, :], rand_mode)
+            batch_input_high[patch_id, :, :, :] = data_augmentation(train_high_data[image_id][x : x+patch_size, y : y+patch_size, :], rand_mode)
+            image_id = (image_id + 1) % len(train_low_data)
+            if image_id == 0:
+                tmp = list(zip(train_low_data, train_high_data))
+                random.shuffle(tmp)
+                train_low_data, train_high_data  = zip(*tmp)
+
+        _, loss = sess.run([train_op, train_loss], feed_dict={input_low: batch_input_low, \
+                                                              input_high: batch_input_high, \
+                                                              lr: learning_rate})
+        print("%s Epoch: [%2d] [%4d/%4d] time: %4.4f, loss: %.6f" \
+              % (train_phase, epoch + 1, batch_id + 1, numBatch, time.time() - start_time, loss))
+        iter_num += 1
+    if (epoch + 1) % eval_every_epoch == 0:
+        print("[*] Evaluating for phase %s / epoch %d..." % (train_phase, epoch + 1))
+        for idx in range(len(eval_low_data)):
+            input_low_eval = np.expand_dims(eval_low_data[idx], axis=0)
+            result_1, result_2 = sess.run([output_R_low, output_I_low], feed_dict={input_low: input_low_eval})
+            save_images(os.path.join(sample_dir, 'low_%d_%d.png' % ( idx + 1, epoch + 1)), result_1, result_2)
+        for idx in range(len(eval_high_data)):
+            input_high_eval = np.expand_dims(eval_high_data[idx], axis=0)
+            result_11, result_22 = sess.run([output_R_high, output_I_high], feed_dict={input_high: input_high_eval})
+            save_images(os.path.join(sample_dir, 'high_%d_%d.png' % ( idx + 1, epoch + 1)), result_11, result_22)
+         
+    saver.save(sess, checkpoint_dir + 'model.ckpt')
+
+print("[*] Finish training for phase %s." % train_phase)

evaluate.py

@@ -0,0 +1,114 @@
+# coding: utf-8
+from __future__ import print_function
+import os
+import time
+import random
+from PIL import Image
+import tensorflow as tf
+import numpy as np
+from utils import *
+from model import *
+from glob import glob
+from skimage import color,filters
+
+sess = tf.Session()
+
+input_decom = tf.placeholder(tf.float32, [None, None, None, 3], name='input_decom')
+input_low_r = tf.placeholder(tf.float32, [None, None, None, 3], name='input_low_r')
+input_low_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i')
+input_high_r = tf.placeholder(tf.float32, [None, None, None, 3], name='input_high_r')
+input_high_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_high_i')
+input_low_i_ratio = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i_ratio')
+
+[R_decom, I_decom] = DecomNet_simple(input_decom)
+decom_output_R = R_decom
+decom_output_I = I_decom
+output_r = Restoration_net(input_low_r, input_low_i)
+output_i = Illumination_adjust_net(input_low_i, input_low_i_ratio)
+
+var_Decom = [var for var in tf.trainable_variables() if 'DecomNet' in var.name]
+var_adjust = [var for var in tf.trainable_variables() if 'Illumination_adjust_net' in var.name]
+var_restoration = [var for var in tf.trainable_variables() if 'Restoration_net' in var.name]
+
+saver_Decom = tf.train.Saver(var_list = var_Decom)
+saver_adjust = tf.train.Saver(var_list=var_adjust)
+saver_restoration = tf.train.Saver(var_list=var_restoration)
+
+decom_checkpoint_dir ='./checkpoint/decom_net_train/'
+ckpt_pre=tf.train.get_checkpoint_state(decom_checkpoint_dir)
+if ckpt_pre:
+    print('loaded '+ckpt_pre.model_checkpoint_path)
+    saver_Decom.restore(sess,ckpt_pre.model_checkpoint_path)
+else:
+    print('No decomnet checkpoint!')
+
+checkpoint_dir_adjust = './checkpoint/illumination_adjust_net_train/'
+ckpt_adjust=tf.train.get_checkpoint_state(checkpoint_dir_adjust)
+if ckpt_adjust:
+    print('loaded '+ckpt_adjust.model_checkpoint_path)
+    saver_adjust.restore(sess,ckpt_adjust.model_checkpoint_path)
+else:
+    print("No adjust pre model!")
+
+checkpoint_dir_restoration = './checkpoint/Restoration_net_train/'
+ckpt=tf.train.get_checkpoint_state(checkpoint_dir_restoration)
+if ckpt:
+    print('loaded '+ckpt.model_checkpoint_path)
+    saver_restoration.restore(sess,ckpt.model_checkpoint_path)
+else:
+    print("No restoration pre model!")
+
+###load eval data
+eval_low_data = []
+eval_img_name =[]
+eval_low_data_name = glob('./test/*')
+eval_low_data_name.sort()
+for idx in range(len(eval_low_data_name)):
+    [_, name] = os.path.split(eval_low_data_name[idx])
+    suffix = name[name.find('.') + 1:]
+    name = name[:name.find('.')]
+    eval_img_name.append(name)
+    eval_low_im = load_images(eval_low_data_name[idx])
+    eval_low_data.append(eval_low_im)
+    print(eval_low_im.shape)
+
+sample_dir = './results/test/'
+if not os.path.isdir(sample_dir):
+    os.makedirs(sample_dir)
+
+print("Start evalating!")
+start_time = time.time()
+for idx in range(len(eval_low_data)):
+    print(idx)
+    name = eval_img_name[idx]
+    input_low = eval_low_data[idx]
+    input_low_eval = np.expand_dims(input_low, axis=0)
+    h, w, _ = input_low.shape
+
+    decom_r_low, decom_i_low = sess.run([decom_output_R, decom_output_I], feed_dict={input_decom: input_low_eval})
+    
+    restoration_r = sess.run(output_r, feed_dict={input_low_r: decom_r_low, input_low_i: decom_i_low})
+### change the ratio to get different exposure level, the value can be 0-5.0
+    ratio = 5.0
+    i_low_data_ratio = np.ones([h, w])*(ratio)
+    i_low_ratio_expand = np.expand_dims(i_low_data_ratio , axis =2)
+    i_low_ratio_expand2 = np.expand_dims(i_low_ratio_expand, axis=0)
+    adjust_i = sess.run(output_i, feed_dict={input_low_i: decom_i_low, input_low_i_ratio: i_low_ratio_expand2})
+
+    #The restoration result can find more details from very dark regions, however, it will restore the very dark regions
+#with gray colors, we use the following operator to alleviate this weakness.  
+    decom_r_sq = np.squeeze(decom_r_low)
+    r_gray = color.rgb2gray(decom_r_sq)
+    r_gray_gaussion = filters.gaussian(r_gray, 3)
+    low_i =  np.minimum((r_gray_gaussion*2)**0.5,1)
+    low_i_expand_0 = np.expand_dims(low_i, axis = 0)
+    low_i_expand_3 = np.expand_dims(low_i_expand_0, axis = 3)
+    result_denoise = restoration_r*low_i_expand_3
+    fusion4 = result_denoise*adjust_i
+    
+    #fusion = restoration_r*adjust_i
+# fuse with the original input to avoid over-exposure
+    fusion2 = decom_i_low*input_low_eval + (1-decom_i_low)*fusion4
+    #print(fusion2.shape)
+    save_images(os.path.join(sample_dir, '%s_kindle.png' % (name)), fusion2)
+    

evaluate_LOLdataset.py

@@ -0,0 +1,110 @@
+# coding: utf-8
+from __future__ import print_function
+import os
+import time
+import random
+from PIL import Image
+import tensorflow as tf
+import numpy as np
+from utils import *
+from model import *
+from glob import glob
+
+sess = tf.Session()
+
+input_decom = tf.placeholder(tf.float32, [None, None, None, 3], name='input_decom')
+input_low_r = tf.placeholder(tf.float32, [None, None, None, 3], name='input_low_r')
+input_low_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i')
+input_high_r = tf.placeholder(tf.float32, [None, None, None, 3], name='input_high_r')
+input_high_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_high_i')
+input_low_i_ratio = tf.placeholder(tf.float32, [None, None, None, 1], name='input_low_i_ratio')
+
+[R_decom, I_decom] = DecomNet_simple(input_decom)
+decom_output_R = R_decom
+decom_output_I = I_decom
+output_r = Restoration_net(input_low_r, input_low_i)
+output_i = Illumination_adjust_net(input_low_i, input_low_i_ratio)
+
+var_Decom = [var for var in tf.trainable_variables() if 'DecomNet' in var.name]
+var_adjust = [var for var in tf.trainable_variables() if 'Illumination_adjust_net' in var.name]
+var_restoration = [var for var in tf.trainable_variables() if 'Restoration_net' in var.name]
+
+saver_Decom = tf.train.Saver(var_list = var_Decom)
+saver_adjust = tf.train.Saver(var_list=var_adjust)
+saver_restoration = tf.train.Saver(var_list=var_restoration)
+
+decom_checkpoint_dir ='./checkpoint/decom_net_train/'
+ckpt_pre=tf.train.get_checkpoint_state(decom_checkpoint_dir)
+if ckpt_pre:
+    print('loaded '+ckpt_pre.model_checkpoint_path)
+    saver_Decom.restore(sess,ckpt_pre.model_checkpoint_path)
+else:
+    print('No decomnet checkpoint!')
+
+checkpoint_dir_adjust = './checkpoint/illumination_adjust_net_train/'
+ckpt_adjust=tf.train.get_checkpoint_state(checkpoint_dir_adjust)
+if ckpt_adjust:
+    print('loaded '+ckpt_adjust.model_checkpoint_path)
+    saver_adjust.restore(sess,ckpt_adjust.model_checkpoint_path)
+else:
+    print("No adjust pre model!")
+
+checkpoint_dir_restoration = './checkpoint/Restoration_net_train/'
+ckpt=tf.train.get_checkpoint_state(checkpoint_dir_restoration)
+if ckpt:
+    print('loaded '+ckpt.model_checkpoint_path)
+    saver_restoration.restore(sess,ckpt.model_checkpoint_path)
+else:
+    print("No restoration pre model!")
+
+###load eval data
+eval_low_data = []
+eval_img_name =[]
+eval_low_data_name = glob('./LOLdataset/eval15/low/*.png')
+eval_low_data_name.sort()
+for idx in range(len(eval_low_data_name)):
+    [_, name] = os.path.split(eval_low_data_name[idx])
+    suffix = name[name.find('.') + 1:]
+    name = name[:name.find('.')]
+    eval_img_name.append(name)
+    eval_low_im = load_images(eval_low_data_name[idx])
+    eval_low_data.append(eval_low_im)
+    print(eval_low_im.shape)
+# To get better results, the illumination adjustment ratio is computed based on the decom_i_high, so we also need the high data.
+eval_high_data = []
+eval_high_data_name = glob('./LOLdataset/eval15/high/*.png')
+eval_high_data_name.sort()
+for idx in range(len(eval_high_data_name)):
+    eval_high_im = load_images(eval_high_data_name[idx])
+    eval_high_data.append(eval_high_im)
+
+sample_dir = './results/LOLdataset_eval15/'
+if not os.path.isdir(sample_dir):
+    os.makedirs(sample_dir)
+
+print("Start evalating!")
+start_time = time.time()
+for idx in range(len(eval_low_data)):
+    print(idx)
+    name = eval_img_name[idx]
+    input_low = eval_low_data[idx]
+    input_low_eval = np.expand_dims(input_low, axis=0)
+    input_high = eval_high_data[idx]
+    input_high_eval = np.expand_dims(input_high, axis=0)
+    h, w, _ = input_low.shape
+
+    decom_r_low, decom_i_low = sess.run([decom_output_R, decom_output_I], feed_dict={input_decom: input_low_eval})
+    decom_r_high, decom_i_high = sess.run([decom_output_R, decom_output_I], feed_dict={input_decom: input_high_eval})
+    
+    restoration_r = sess.run(output_r, feed_dict={input_low_r: decom_r_low, input_low_i: decom_i_low})
+
+    ratio = np.mean(((decom_i_high))/(decom_i_low+0.0001))
+    
+    i_low_data_ratio = np.ones([h, w])*(ratio)
+    i_low_ratio_expand = np.expand_dims(i_low_data_ratio , axis =2)
+    i_low_ratio_expand2 = np.expand_dims(i_low_ratio_expand, axis=0)
+
+    adjust_i = sess.run(output_i, feed_dict={input_low_i: decom_i_low, input_low_i_ratio: i_low_ratio_expand2})
+    fusion = restoration_r*adjust_i
+    save_images(os.path.join(sample_dir, '%s_kindle.png' % (name)), fusion)
+