Deep Variational Instance Segmentation

    ████████║   ██         ██  ██████████    █████████║
    ██      █║   █         █      ║██║       █║
    ██       █║   █       █       ║██║       █████████║
    ██       █║    █     █        ║██║              ██║
    ██      █║      █   █         ║██║              ██║
    ███████║        █████      ██████████    █████████

A simple, fully convolutional model for real-time instance segmentation. This is the code for our papers:

• Deep Variational Instance Segmentation

The implementation of backbone network is based on repository: Yolact-github

Installation

• Clone this repository and enter it:

  git clone https://github.com/jia2lin3yuan1/2020-instanceSeg.git $PRJ_NAME
  cd  $PRJ_NAME

• Set up the environment using one of the following methods:
  • Using Anaconda
    • Run conda env create -f environment.yml
  • Manually with pip
    • Set up a Python3 environment (e.g., using virtenv).
    • Install Pytorch 1.0.1 (or higher) and TorchVision.
    • Install some other packages:

      # Cython needs to be installed before pycocotools
      pip install cython
      pip install opencv-python pillow pycocotools matplotlib 
      pip install scikit-image
      pip install scipy==1.2.0

• If you'd like to train DVIS, download the COCO dataset and the 2014/2017 annotations. Note that this script will take a while and dump 21gb of files into ./data/coco.

  sh data/scripts/COCO.sh

• If you'd like to evaluate DVIS on test-dev, download test-dev with this script.

  sh data/scripts/COCO_test.sh

• Install pymeanshift following instruction on Install

Train on your own dataset:

• You could edit the config_xx.py in data/ to customize the network setting and dataset setting.
• You could run with specific the arguments on shell command:

   python train.py --config=plus_resnet50_config_550 --resume=PATH/TO/YOUR/FILE --start_iter=0 --exp_name=dummy     

• Or, you could customize the json script in exp_scripts/, and run with:

   python train.py --scripts=exp_scripts/xxx.json