something.py

# -*- coding: utf-8 -*-
from __future__ import unicode_literals
import tensorflow as tf
import numpy as np
import glob
import sys
from models import *
import yaml
import os
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import utils

####### Reading Hyperparameters #####
with open("config.yaml") as file:
	data = yaml.load(file)
	training_params = data['training_params']
	learning_rate = float(training_params['learning_rate'])
	batch_size = int(training_params['batch_size'])
	epoch_size = int(training_params['epochs'])
	data_path = training_params['data_path']
	model_params= data['model_params']
	descrip = model_params['descrip']
	log_dir = model_params['log_dir']
	mode = model_params['mode']
	if len(descrip)==0:
		raise ValueError, "Please give a proper description of the model you are training."


######## Making Directory #########
model_path = log_dir+sys.argv[1]
print "Model Path: ", model_path
if not os.path.exists(model_path):
	os.makedirs(model_path)
	os.makedirs(model_path+"/results")
	os.makedirs(model_path+"/tf_graph")
	os.makedirs(model_path+"/saved_model")


######### Loading Data ###########
train_img1 = 1/255.0*np.load(data_path+"train_haze_clear.npy") # First image of each pair is hazy image and second image is clear images
train_img2 = 1/255.0*np.load(data_path+"train_trans.npy") 
val_img1 = 1/255.0*np.load(data_path+"val_haze_clear.npy")
val_img2 = 1/255.0*np.load(data_path+"val_trans.npy")	   
print "Data Loaded"

def main():
	nnet = MSCNN(model_path)
	if mode=='train':
		os.system('cp config.yaml '+model_path+'/config.yaml')
		os.system('cp models.py '+model_path+'/model.py')
		nnet.build_model()
		print "Model Build......"
		nnet.train_model([train_img1, train_img2], [val_img1, val_img2], learning_rate, batch_size, epoch_size)
	else:
		predict_maps, predict_clear = nnet.test(train_img1[:,0,:,:,:], batch_size)
		for i in range(train_img2.shape[0]):
			pair = np.hstack((train_img2[i], predict_maps[i]))
			pair2 = np.hstack((train_img1[i,0,:,:,:],train_img1[i,1,:,:,:], predict_clear[i]))
			cv2.imwrite(model_path+"/results/train/"+str(i)+"_trans.jpg", 255.0*pair)
			cv2.imwrite(model_path+"/results/train/"+str(i)+"_clear.jpg", 255.0*pair2)

if __name__ == "__main__":
	main()