This repository contains:

• the PyTorch implementation of ShapeProp.
• the Classwise semi-supervision (COCO's VOC->Non-VOC) demo.

Please follow the instruction below to install it and run the experiment demo.

• Linux (tested on ubuntu 16.04LTS)
• NVIDIA GPU + CUDA CuDNN (tested on 8x GTX 2080 Ti)
• COCO 2017 Dataset (download and unzip)
• Please use PyTorch1.1 + Apex(#1564802) to avoid compilation errors

0. Create a conda environment:

   conda create --name ShapeProp -y
   conda activate ShapeProp

1. Clone this repo:

   # git version must be greater than 1.9.10
   git clone https://github.com/ucbdrive/ShapeProp.git
   cd ShapeProp
   export DIR=$(pwd)

2. Install dependencies via a single command bash $DIR/scripts/install.sh or do it manually as follows:

   # Python
   conda install -y ipython pip
   # PyTorch
   conda install -y pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=10.0 -c pytorch
   # Install deps
   pip install ninja yacs cython matplotlib tqdm opencv-python
   rm -r libs
   mkdir libs
   # COCOAPI
   cd $DIR/libs
   git clone https://github.com/cocodataset/cocoapi.git
   cd cocoapi/PythonAPI
   python setup.py build_ext install
   # APEX
   cd $DIR/libs
   git clone https://github.com/NVIDIA/apex.git
   cd apex
   python setup.py install --cuda_ext --cpp_ext
   # ShapeProp
   cd $DIR
   python setup.py build develop
   

3. Prepare dataset:

   cd $DIR
   mkdir datasets
   ln -s PATH_TO_YOUR_COCO_DATASET datasets/coco
   bash scripts/prepare_data.sh

4. Run the classwise semi-supervision demo:

   cd $DIR
   # Mask R-CNN w/ ShapeProp
   bash scripts/train_shapeprop.sh
   # Mask R-CNN
   bash scripts/train_baseline.sh

If you use the code in your research, please cite:

@INPROCEEDINGS{Zhou2020ShapeProp,
    author = {Zhou, Yanzhao and Wang, Xin and and Jiao, Jianbin and Darrell, Trevor and Yu, Fisher},
    title = {Learning Saliency Propagation for Semi-supervised Instance Segmentation},
    booktitle = {CVPR},
    year = {2020}
}