Setup
32 x 32 Downsampled MS COCO (eventually use full size 2017 data)
-
sudo rm -R WGAN/inpainting
-
cd WGAN
-
wget http://lisaweb.iro.umontreal.ca/transfert/lisa/datasets/mscoco_inpaiting/inpainting.tar.bz2
-
tar xvjf inpainting.tar.bz2 should have two folders "train2014" and "val2014"
-
cd WGAN/inpainting/
-
wget http://images.cocodataset.org/annotations/annotations_trainval2014.zip
Fist Time Install cocoapi + Create Conda Environment (if not built already)
- cd WGAN
- git clone https://github.com/cocodataset/cocoapi.git
- conda create -n cs231n python=3.6 anaconda
- source activate cs231n
- cd cocoapi/PythonAPI
- make
- sudo make install
- python setup.py install
make sure to activate conda and then reinstall pytorch using
- conda install pytorch torchvision -c pytorch
Afterwards to run notebook (always run)
- conda activate cs231n
it may not work on jupyter labs, you might have to boot up the original way