Requirements

To run this program, you need install at least

• Pytorch
• Numpy (might be automatically install with Pytorch)
• tqdm (for progress bar)

Other things might be require. Install the modules suit you environment.

Original Paper

Pixel Objectness

Dataset

You need to prepare your dataset.
Download the dataset from

• PASCAL VOC dataset
• SBD dataset

Preprocess

First, you need to preprocess the dataset to make binalized mask image.

python misc/create_dataset.py --voc_root PATH_TO/VOCdevkit/ --sbd_root PATH_TO/SBDdataset/benchmark_RELEASE

This will do the all things and, will make train.pkl and val.pkl which are pickled dataset.

Train

First, you need to download the ImageNet pretrained parameter from pytorch official.

wget https://download.pytorch.org/models/vgg16-397923af.pth

then

python train.py

There are some options you can change. See the code or use --help for more detatil.

Prediction

python prediction.py --image_dir folder_that_contains_images/