pranjal-joshi

#include<reg51.h>

#define cmdport P3
#define dataport P2
#define q 10
#define o 1

sbit s1 = (P0^1);
sbit s2 = (P0^2);
sbit s3 = (P0^3);

pranjal-joshi

evaluator = Evaluator()
x = preprocess_img(base_path)
z = load_img(base_path,target_size=(img_nrows,img_ncols))
imsave('iteration_original.png',z)
print "Saving original image.."

for i in range(EPOCHS):
    print "Start iteration: %d" % i
    start_time = time.time()
    x,min_val,info = fmin_l_bfgs_b(evaluator.loss,x.flatten(),fprime=evaluator.grads,maxfun=20)

pranjal-joshi

def eval_loss_and_grads(x):
    x = x.reshape((1,3,img_nrows,img_ncols))
    outs = f_outputs([x])
    loss_value = outs[0]
    if(len(outs[1:])==1):
        grad_values = outs[1].flatten().astype('float64')
    else:
        grad_values = np.array(outs[1:]).flatten().astype('float64')
    return loss_value,grad_values

pranjal-joshi

loss = K.variable(0.0)
layer_features = output_dict['block4_conv2']
b_img_features = layer_features[0,:,:,:]
final_features = layer_features[2,:,:,:]
loss += content_wt * content_loss(b_img_features,final_features)

feature_layers = ['block1_conv1','block2_conv1','block3_conv1','block4_conv1','block5_conv1']

for layer in feature_layers:
    layer_features = output_dict[layer]

pranjal-joshi

def content_loss(base,final):
    return K.sum(K.square(final-base))		# square error

def gram_matrix(x):
    features = K.batch_flatten(x)		#flatten -> converts multidimension matrix to single dimensional array
    gram = K.dot(features,K.transpose(features))
    return gram

def style_loss(style,final):
    s = gram_matrix(style)

pranjal-joshi

b_img = K.variable(preprocess_img(base_path))
r_img = K.variable(preprocess_img(ref_path))
final_img = K.placeholder((1,3,img_nrows,img_ncols))
input_tensor = K.concatenate([b_img, r_img, final_img],axis=0)

model = vgg16.VGG16(input_tensor=input_tensor,weights='imagenet',include_top=False)

output_dict = dict([(layer.name,layer.output) for layer in model.layers])

pranjal-joshi

def preprocess_img(path):
    print "preprocess: " + path
    i = load_img(path,target_size=(img_nrows,img_ncols))
    i = img_to_array(i)
    i = np.expand_dims(i,axis=0)
    i = vgg16.preprocess_input(i)
    return i

def deprocess_img(i):
    i = i.reshape((3,img_nrows,img_ncols))

pranjal-joshi

#!/usr/bin/python

import os
from scipy.optimize import fmin_l_bfgs_b
from scipy.misc import imsave, imread, imresize
from keras.applications import vgg16
from keras.preprocessing.image import load_img, img_to_array
from keras import backend as K
import time
import numpy as np

pranjal-joshi

[global]
floatX = float32
device = gpu

[lib]
cnmem = 0

[dnn]
enabled = auto

pranjal-joshi

{
    "image_dim_ordering": "th", 
    "epsilon": 1e-07, 
    "floatx": "float32", 
    "backend": "theano"
}