Deeplearning Class Inpainting dataset, http://ift6266h17.wordpress.com/ *

** Dataset

The inpainting dataset is a downsample version of the MSCOCO dataset (http://mscoco.org/dataset/#overview)

The original images in the MSCOCO dataset are high resolution: roughly 500×500 pixels. While this will give the most interesting results, it would likely be too computationally intensive for a class project (and indeed, for most research projects). Many successful generative modeling papers still evaluate on 32×32 images. In fact, there can still be a lot of detail in a 32×32 image!

Some of the images might be in grayscale instead of RGB, you can just skip those images.

** Files http://lisaweb.iro.umontreal.ca/transfert/lisa/datasets/mscoco_inpaiting/

In particular the archive inpainting.tar.bz2 contains:

• train2014: a directory composed by 82782 training images
• val2014:a directory composed by 40504 validations images
• dict_key_imgID_value_caps_train_and_valid.pkl: a pickled python dictionary containing the captions associated to the train/valid images
• worddict.pkl: a pickled dictionary of the different words composing the captions.

To extract the archive: tar xjvf inpainting.tar.bz2

** Examples

We also provides the python script examples.py Code used to downsample the images, and a function to visualize the batch are in examples.py We also show how to construct the input/target from the images and how to visualize the datas,

Please refer to the class website, http://ift6266h17.wordpress.com/project-description/, for more details.

** Useful commands

Saving dependencies :
pip freeze > requirements.txt

Installing dependencies :
pip instrall -r requirements.txt

Using virtualenv as default commands (deactivate to reverse):
source env/bin/activate

On Windows the easiest way is to use Anaconda python 2.7 and install dependencies such as Theanos like this:
conda install theano pygpu
Anaconda also allows to create an environment to encapsulate all dependencies for this project

You can activate the Anaconda environment with the following command activate deeplearning