diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..13471e2
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,6 @@
+*.caffemodel
+.idea/
+*.png
+*.jpg
+*.dump
+*.pyc
\ No newline at end of file
diff --git a/README.md b/README.md
index 73bd9a2..50a9fc0 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,13 @@ neural_style_synthesizer
 INSTALL
 ---------------
 
+The model files of neural networks are not contained in this repository.
+You can get them from [nin_imagenet.caffemodel](https://gist.github.com/mavenlin/d802a5849de39225bcc6) and [VGG_ILSVRC_16_layers.caffemodel](https://gist.github.com/ksimonyan/211839e770f7b538e2d8#file-readme-md).
+
+Dependent libraries are installed with the following script.
+
 ```
+pip install numpy
 pip install -r requirements.txt
 ```
 
@@ -19,6 +25,7 @@ with CPU
 
 ```
 python bin/convert_image_multi.py \
+  --modelpath=./VGG_ILSVRC_16_layers.caffemodel \
   --iteration=100 \
   --gpu=-1 \
   --xsplit=1 --ysplit=1 --resize=300 \
@@ -31,6 +38,7 @@ with GPU
 
 ```
 python bin/convert_image_multi.py \
+  --modelpath=./VGG_ILSVRC_16_layers.caffemodel \
   --iteration=100 \
   --gpu=0 \
   --xsplit=1 --ysplit=1 --resize=300 \
@@ -46,6 +54,7 @@ Split style image to 2x2
 
 ```
 python bin/convert_image_multi.py \
+  --modelpath=./VGG_ILSVRC_16_layers.caffemodel \
   --iteration=100 \
   --gpu=0 \
   --xsplit=2 --ysplit=2 --resize=300 \
diff --git a/bin/convert_image_multi.py b/bin/convert_image_multi.py
index 7dd2864..d35df35 100644
--- a/bin/convert_image_multi.py
+++ b/bin/convert_image_multi.py
@@ -22,6 +22,7 @@
 parser.add_argument("--debug_span", type=int, default=100)
 parser.add_argument("--average_pooling", action="store_true")
 parser.add_argument("--model", default="vgg")
+parser.add_argument("--modelpath")
 parser.add_argument("--random_init", action="store_true")
 parser.add_argument("--init_image", default=None)
 args = parser.parse_args()
@@ -44,7 +45,7 @@
 init_img = neural_art.utility.load_image(args.init_image)
 init_img = neural_art.utility.resize_img(init_img, args.resize)
 
-model = neural_art.utility.load_nn(args.model)
+model = neural_art.utility.load_nn(args.model, modelpath=args.modelpath)
 converter = neural_art.image_converters.MultiReferenceImageConverter(texture_imgs, gpu=args.gpu, content_weight=args.content_weight, texture_weight=1, model=model, average_pooling=args.average_pooling)
 
 if args.debug:
diff --git a/neural_art/__init__.pyc b/neural_art/__init__.pyc
deleted file mode 100644
index a724036..0000000
Binary files a/neural_art/__init__.pyc and /dev/null differ
diff --git a/neural_art/image_converters/__init__.pyc b/neural_art/image_converters/__init__.pyc
deleted file mode 100644
index 3d6634a..0000000
Binary files a/neural_art/image_converters/__init__.pyc and /dev/null differ
diff --git a/neural_art/image_converters/image_converter.pyc b/neural_art/image_converters/image_converter.pyc
deleted file mode 100644
index 2dc6720..0000000
Binary files a/neural_art/image_converters/image_converter.pyc and /dev/null differ
diff --git a/neural_art/image_converters/multi_reference_image_converter.py b/neural_art/image_converters/multi_reference_image_converter.py
index 809bdad..81a5c18 100644
--- a/neural_art/image_converters/multi_reference_image_converter.py
+++ b/neural_art/image_converters/multi_reference_image_converter.py
@@ -5,7 +5,6 @@
 import chainer.functions
 import neural_art
 import numpy
-import scipy.optimize
 import image_converter
 import openopt
 
diff --git a/neural_art/image_converters/multi_reference_image_converter.pyc b/neural_art/image_converters/multi_reference_image_converter.pyc
deleted file mode 100644
index aa1486a..0000000
Binary files a/neural_art/image_converters/multi_reference_image_converter.pyc and /dev/null differ
diff --git a/neural_art/models/__init__.pyc b/neural_art/models/__init__.pyc
deleted file mode 100644
index 82c1af3..0000000
Binary files a/neural_art/models/__init__.pyc and /dev/null differ
diff --git a/neural_art/models/vgg.pyc b/neural_art/models/vgg.pyc
deleted file mode 100644
index 2d0c091..0000000
Binary files a/neural_art/models/vgg.pyc and /dev/null differ
diff --git a/neural_art/utility.py b/neural_art/utility.py
index b562a64..6a14cc7 100644
--- a/neural_art/utility.py
+++ b/neural_art/utility.py
@@ -29,17 +29,20 @@ def resize_img(img, max_length):
         new_h = max_length*orig_h/orig_w
     return img.resize((new_w,new_h))
 
-def load_nn(modelname):
+def load_nn(modelname, modelpath = None):
     cachepath = "{}.dump".format(modelname)
     if os.path.exists(cachepath):
         nn = pickle.load(open(cachepath))
     else:
         if modelname == 'vgg':
-            nn = neural_art.models.VGG("VGG_ILSVRC_16_layers.caffemodel", no_padding=False)
+            if modelpath is None: modelpath = "VGG_ILSVRC_16_layers.caffemodel"
+            nn = neural_art.models.VGG(modelpath, no_padding=False)
         elif modelname == 'vgg_nopad':
-            nn = neural_art.models.VGG("VGG_ILSVRC_16_layers.caffemodel", no_padding=True)
+            if modelpath is None: modelpath = "VGG_ILSVRC_16_layers.caffemodel"
+            nn = neural_art.models.VGG(modelpath, no_padding=True)
         elif modelname == 'nin':
-            nn = neural_art.models.NIN("nin_imagenet.caffemodel")
+            if modelpath is None: modelpath = "nin_imagenet.caffemodel"
+            nn = neural_art.models.NIN(modelpath)
         else:
             print 'invalid model name.'
             exit(1)
diff --git a/neural_art/utility.pyc b/neural_art/utility.pyc
deleted file mode 100644
index 96dfec5..0000000
Binary files a/neural_art/utility.pyc and /dev/null differ
diff --git a/neural_art/utility_old.pyc b/neural_art/utility_old.pyc
deleted file mode 100644
index bd6a2e1..0000000
Binary files a/neural_art/utility_old.pyc and /dev/null differ
diff --git a/requirements.txt b/requirements.txt
index 2af8018..ed4310d 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,2 +1,5 @@
+numpy
 openopt
 cvxopt
+chainer
+pillow