luhc15/chainer-fcis

Fully Convolutional Instance-aware Semantic Segmentation

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.
• Large GPU memory around 10GB is required (Titan X, GeForce 1080Ti).

Installation

# Requirement installation
# I recommend to use anacoda.

conda create -n fcis python=2.7
conda install -c menpo opencv
pip install cupy

# Installation

git clone https://github.com/knorth55/chainer-fcis.git
cd chainer-fcis
pip install -e .

Inference

Pretrained models can be dowloaded here.

Inference can be done as below;

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