-
Notifications
You must be signed in to change notification settings - Fork 1
/
README
53 lines (35 loc) · 2.08 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
Image Colorization
== Notes ==
Note1 : Some architectures need very large memory
(We used AWS EC2 P2 instance that has 61 GB memory to do the experiments)
Note2 : Our results can be view from google drive:
https://drive.google.com/open?id=0BzxDU3VAcOkQck9rYThkd3JmQVE
Execution time : Run at least 30 minutes (1 hour for some larger architecture) for each code
Results : View the result image in corresponding folder namde predict_output_(architecture)
== Dependencies ==
OS: Ubuntu 16.04
Software: python
Deep learning framework: Theano, Keras
Dependencies: openCV
== Dataset Download Links: ==
Since we rearranged the dataset, we uploaded the dataset to the following links
1.fruitdata : https://drive.google.com/open?id=0BzxDU3VAcOkQUVRhTlJ6SXlIYnc
2.combined : https://drive.google.com/open?id=0BzxDU3VAcOkQTnpmZjA2aGszWGM
After download the dataset, fruitdata should be put into the fruit_colorizer folder,
and combined should be put into landscape_colorizer folder
== Run Code ==
1.For training and doing predicting for Fruit dataset:
$ cd fruit_colorizer
$ colorization_auto_encoder.py # Use deep auto-encoder architecture
$ colorization_auto_encoder_shallow.py # Use shallow auto-encoder architecture
2.For training and doing predicting for Open Country(Landscape) dataset:
$ cd landscape_colorizer
$ python colorization_use_pretrained_conv_layers.py # Use our pre-trained conv-layer and Dense layer.
# This is our baseline model
$ python colorization_concate_pretrained_conv_layers.py # Use concat conv layer architecture
$ python colorization_auto_encoder.py # Use auto-encoder architecture
$ python colorization_use_vgg.py # Use vgg as feature extractor
$ python colorization_use_concate_vgg.py # Use concat vgg conv layers architecture
3. For comparison between color images and gray scale images testing on a VGG-16 fine-tuned network. The result shows loss and accuracy.
$ cd comparison
$ python vgg16_cifar_RGB_gray_comparison.py