This project uses Keras and Python to convert a grayscale image to color without any additional information.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
comparison Add files via upload Oct 17, 2017
fruit_colorizer Add files via upload Oct 17, 2017
LICENSE Rename LICENSE.txt to LICENSE Oct 17, 2017
README Update README Oct 17, 2017


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:

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 :

2.combined :

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

$             # Use deep auto-encoder architecture 
$     # Use shallow auto-encoder architecture 

2.For training and doing predicting for Open Country(Landscape) dataset:
$ cd landscape_colorizer

$ python          # Use our pre-trained conv-layer and Dense layer.
                                                             # This is our baseline model

$ python      # Use concat conv layer architecture
$ python                        # Use auto-encoder architecture
$ python                             # Use vgg as feature extractor
$ python                     # 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