reflectance_restoration_net_train.py

# 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

training = tf.placeholder_with_default(False, shape=(), name='training')
batch_size = 10
patch_size = 48

config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess=tf.Session(config=config)
#decomnet input
input_all = tf.placeholder(tf.float32, [None, None, None, 3], name='input_all')

#restoration net input
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 = tf.placeholder(tf.float32, [None, None, None, 3], name='input_high')
input_high_i = tf.placeholder(tf.float32, [None, None, None, 1], name='input_high_i')

[R_decom, I_decom] = DecomNet(input_all)
output_r = Restoration_net(input_low_r, input_low_i, training)

#network output
output_R_all = R_decom
output_I_all = I_decom

#define loss
#ssim loss
output_r_1 = output_r[:,:,:,0:1]
input_high_1 = input_high[:,:,:,0:1]
ssim_r_1 = tf_ssim(output_r_1, input_high_1)
output_r_2 = output_r[:,:,:,1:2]
input_high_2 = input_high[:,:,:,1:2]
ssim_r_2= tf_ssim(output_r_2, input_high_2)
output_r_3 = output_r[:,:,:,2:3]
input_high_3 = input_high[:,:,:,2:3]
ssim_r_3 = tf_ssim(output_r_3, input_high_3)
ssim_r = (ssim_r_1 + ssim_r_2 + ssim_r_3)/3.0
loss_ssim = 1-ssim_r
#mse loss
loss_square = tf.reduce_mean(tf.square(output_r  - input_high))#*(1-input_low_i))# * ( 1 - input_low_r ))#* (1- input_low_i)))
#total loss
loss_restoration =  1*loss_square + 1*loss_ssim

lr = tf.placeholder(tf.float32, name='learning_rate')


global_step = tf.get_variable('global_step', [], dtype=tf.int32, initializer=tf.constant_initializer(0), trainable=False)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
optimizer = tf.train.AdamOptimizer(learning_rate=lr, name='AdamOptimizer')
with tf.control_dependencies(update_ops):
    grads = optimizer.compute_gradients(loss_restoration)
    train_op_restoration = optimizer.apply_gradients(grads, global_step=global_step)

var_Decom = [var for var in tf.trainable_variables() if 'DecomNet' in var.name]
var_restoration = [var for var in tf.trainable_variables() if 'Denoise_Net' in var.name]
g_list = tf.global_variables()
bn_moving_vars = [g for g in g_list if 'moving_mean' in g.name]
bn_moving_vars += [g for g in g_list if 'moving_variance' in g.name]
var_restoration += bn_moving_vars

saver_restoration = tf.train.Saver(var_list=var_restoration)
saver_Decom = tf.train.Saver(var_list = var_Decom)
sess.run(tf.global_variables_initializer())
print("[*] Initialize model successfully...")

eval_low_data = []
eval_high_data = []
eval_low_data_bmp = []

eval_low_data_name =  glob('./LOLdataset/our485/low/*.png')+ glob('./LOLdataset/add_sys/sys_low/*.png')+ glob('./LOLdataset/dark/low/*.png')
eval_low_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_low_data_name_bmp = glob('./LOLdataset/eval15/low/*.png')
eval_low_data_name_bmp.sort()
for idx in range(len(eval_low_data_name_bmp)):
    eval_low_im = load_images(eval_low_data_name_bmp[idx])
    eval_low_data_bmp.append(eval_low_im)
    print(eval_low_im.shape)

eval_high_data_name =  glob('./LOLdataset/our485/high/*.png')+ glob('./LOLdataset//add_sys/sys_high/*.png')+ glob('./LOLdataset/dark/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)
    print(eval_high_im.shape)

pre_checkpoint_dir = './checkpoint/decom_model/'
ckpt_pre=tf.train.get_checkpoint_state(pre_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_checkpoint!')

train_restoration_low_r_data_480 = []
train_restoration_low_i_data_480 = []
train_restoration_high_r_data_480 = []

for idx in range(len(eval_high_data)):

    input_low_eval = np.expand_dims(eval_high_data[idx], axis=0)
    print(idx)
    result_1, result_2 = sess.run([output_R_all, output_I_all], feed_dict={input_all: input_low_eval})
    result_1 = (result_1*0.99)**1.2
    result_1_sq = np.squeeze(result_1)
    result_2_sq = np.squeeze(result_2)
    print(result_1.shape, result_2.shape)
    train_restoration_high_r_data_480.append(result_1_sq)

for idx in range(len(eval_low_data)):
    input_low_eval = np.expand_dims(eval_low_data[idx], axis=0)
    print(idx)
    result_11, result_12 = sess.run([output_R_all, output_I_all], feed_dict={input_all: input_low_eval})
    result_11_sq = np.squeeze(result_11)
    result_12_sq = np.squeeze(result_12)
    print(result_11.shape, result_12.shape)
    train_restoration_low_r_data_480.append(result_11_sq)
    train_restoration_low_i_data_480.append(result_12_sq)

eval_restoration_low_r_data_bmp = []
eval_restoration_low_i_data_bmp = []
for idx in range(len(eval_low_data_bmp)):
    input_low_eval = np.expand_dims(eval_low_data_bmp[idx], axis=0)
    print(idx)
    result_11, result_12 = sess.run([output_R_all, output_I_all], feed_dict={input_all: input_low_eval})
    result_11_sq = np.squeeze(result_11)
    result_12_sq = np.squeeze(result_12)
    print(result_11.shape, result_12.shape)
    eval_restoration_low_r_data_bmp.append(result_11_sq)
    eval_restoration_low_i_data_bmp.append(result_12_sq)

eval_restoration_low_r_data = train_restoration_low_r_data_480[235:240]
eval_restoration_low_i_data = train_restoration_low_i_data_480[235:240]

train_restoration_low_r_data = train_restoration_low_r_data_480[0:234] + train_restoration_low_r_data_480[241:-1]
train_restoration_low_i_data = train_restoration_low_i_data_480[0:234] + train_restoration_low_i_data_480[241:-1]
train_restoration_high_r_data = train_restoration_high_r_data_480[0:234] + train_restoration_high_r_data_480[241:-1]
print(len(train_restoration_high_r_data), len(train_restoration_low_r_data),len(train_restoration_low_i_data))
print(len(eval_restoration_low_r_data),len(eval_restoration_low_i_data))
assert len(train_restoration_high_r_data) == len(train_restoration_low_r_data)
assert len(train_restoration_low_i_data) == len(train_restoration_low_r_data)

print('[*] Number of training data: %d' % len(train_restoration_high_r_data))
global_step = tf.Variable(0, name='global_step', trainable=False)
learning_rate = 0.0001
def lr_schedule(epoch):
    initial_lr = learning_rate
    if epoch<=300:
        lr = initial_lr
    elif epoch<=500:
        lr = initial_lr/2
    elif epoch<=1500:
        lr = initial_lr/4 
    else:
        lr = initial_lr/8 
    return lr

epoch = 2400

sample_dir = './new_restoration_train_results_3/'
if not os.path.isdir(sample_dir):
    os.makedirs(sample_dir)
eval_every_epoch = 150
train_phase = 'restoration'
numBatch = len(train_restoration_low_r_data) // int(batch_size)
train_op = train_op_restoration
train_loss = loss_restoration
saver = saver_restoration

checkpoint_dir = './checkpoint/new_restoration_retrain_3/'
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 restoration 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))
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_r = np.zeros((batch_size, patch_size, patch_size, 3), dtype="float32")
        batch_input_low_i = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")

        batch_input_high = np.zeros((batch_size, patch_size, patch_size, 3), dtype="float32")
        batch_input_high_i = np.zeros((batch_size, patch_size, patch_size, 1), dtype="float32")

        for patch_id in range(batch_size):
            h, w, _ = train_restoration_low_r_data[image_id].shape
            x = random.randint(0, h - patch_size)
            y = random.randint(0, w - patch_size)
            i_low_expand = np.expand_dims(train_restoration_low_i_data[image_id], axis = 2)
            rand_mode = random.randint(0, 7)
            batch_input_low_r[patch_id, :, :, :] = data_augmentation(train_restoration_low_r_data[image_id][x : x+patch_size, y : y+patch_size, :] , rand_mode)#+ np.random.normal(0, 0.1, (patch_size,patch_size,3))  , rand_mode)
            batch_input_low_i[patch_id, :, :, :] = data_augmentation(i_low_expand[x : x+patch_size, y : y+patch_size, :] , rand_mode)#+ np.random.normal(0, 0.1, (patch_size,patch_size,3))  , rand_mode)
            batch_input_high[patch_id, :, :, :] = data_augmentation(train_restoration_high_r_data[image_id][x : x+patch_size, y : y+patch_size, :], rand_mode)

            image_id = (image_id + 1) % len(train_restoration_low_r_data)
            if image_id == 0:
                tmp = list(zip(train_restoration_low_r_data, train_restoration_low_i_data, train_restoration_high_r_data))
                random.shuffle(list(tmp))
                train_restoration_low_r_data, train_restoration_low_i_data, train_restoration_high_r_data = zip(*tmp)

        _, loss = sess.run([train_op, train_loss], feed_dict={input_low_r: batch_input_low_r,input_low_i: batch_input_low_i,\
                                                              input_high: batch_input_high,\
                                                              training: True, lr: lr_schedule(epoch)})
        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_restoration_low_r_data)):
            input_uu_r = eval_restoration_low_r_data[idx] 
            input_low_eval_r = np.expand_dims(input_uu_r, axis=0)
            input_uu_i = eval_restoration_low_i_data[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)
            result_1 = sess.run(output_r, feed_dict={input_low_r: input_low_eval_r, input_low_i: input_low_eval_ii,training: False})
            save_images(os.path.join(sample_dir, 'eval_%d_%d.png' % ( idx + 101, epoch + 1)), result_1)

        for idx in range(len(eval_restoration_low_r_data_bmp)):
            input_uu_r = eval_restoration_low_r_data_bmp[idx] 
            input_low_eval_r = np.expand_dims(input_uu_r, axis=0)
            input_uu_i = eval_restoration_low_i_data_bmp[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)
            result_1 = sess.run(output_r, feed_dict={input_low_r: input_low_eval_r, training: False,\
                                                          input_low_i: input_low_eval_ii})
            save_images(os.path.join(sample_dir, 'eval_bmp_%d_%d.png' % ( idx + 101, epoch + 1)), result_1)
        
        global_step = epoch
        saver.save(sess, checkpoint_dir + 'model.ckpt', global_step=global_step)

print("[*] Finish training for phase %s." % train_phase)