This project has moved to ChainerCV.

chainer-fcis - FCIS

This is Chainer implementation of Fully Convolutional Instance-aware Semantic Segmentation.

Original Mxnet repository is msracver/FCIS.

Requirement

• CuPy
• Chainer
• ChainerCV
• OpenCV2

Additional Requirement

• For COCO Dataset class

  • Cython
  • pycocotools

• For COCO Training

  • OpenMPI
  • nccl
  • ChainerMN

Notification

• Only GPU implementation, No CPU implementation yet.

TODO

• VOC
  • [ ] Reproduce original repo training accuracy
• COCO
  • [ ] Reproduce original repo training accuracy
  • [ ] Refine evaluation code

Installation

We recommend to use Anacoda.

# Requirement installation
conda create -n fcis python=2.7
conda install -c menpo opencv
source activate fcis
pip install cupy

# Installation
pip install fcis

Inference

cd examples/coco/
python demo.py

Above is our implementation output, and below is original.

Training

cd examples/voc/
python train.py

LICENSE

MIT LICENSE

Powered by DL HACKS