Home

Awesome

FastMask

FastMask: Segment Multi-scale Object Candidates in One Shot

(Published as a spotlight paper at CVPR 2017 in Honolulu, Hawaii)

Hexiang Hu*, Shiyi Lan*, Yuning Jiang, Zhimin Cao, Fei Sha (*Equal contribution. Work was done during their internships at Megvii Inc.)

If you are using code or other related resources from this repository, please cite the following paper:

@article{hu2016fastmask,
  title={FastMask: Segment Multi-scale Object Candidates in One Shot},
  author={Hu, Hexiang and Lan, Shiyi and Jiang, Yuning and Cao, Zhimin and Sha, Fei},
  journal={arXiv preprint arXiv:1612.08843},
  year={2016}
}

Requirements and Dependencies

Quick Start

Step in common

We highly recommend you to use anaconda2 on ubuntu 14.04, which is our main experimental environment.

For ubuntu 14.04

sudo apt-get update
sudo apt-get install python-opencv python-pip

and then install cocoapi see COCOApi

git clone --recursive https://github.com/voidrank/FastMask
cd FastMask
mkdir params results
pip install -r requirements.txt
cd caffe-fm
make pycaffe -j 4
cd ..

Demonstrate

Download parameters of pretrained models

Download final model and save it in ./params

Image Demo

python image_demo.py [gpu_id] [model] [input_image_path] [--init_weights=weights] [--threshold=0.9]

This instruction will segment the image at [input_image_path] with models/[model].test.prototxt and params/[weights].caffemodel

There is an example: (suppose that the input image named input.jpg is at ./)

python image_demo.py 0 fm-res39 input.jpg --init_weights=fm-res39_final_params.caffemodel

Video Demo

python video_demo.py [gpu_id] [model] [input_video_path] [output_video_path] [--init_weights=weights] [--threshold=0.9]

Training

Set Up Redis Server for Multiprocess Communication

nohup redis-server redis.conf

Download COCO

cd data
wget http://msvocds.blob.core.windows.net/coco2014/train2014.zip http://msvocds.blob.core.windows.net/annotations-1-0-3/instances_train-val2014.zip http://msvocds.blob.core.windows.net/coco2014/val2014.zip
unzip train.zip instances_train_val2014.zip val2014.zip
cd ..

Download pretrained model on imagenet

Download ResNet-50-model.caffemodel and save it in ./params

Training

python train.py [gpu_id] [model] [--init_weights=ResNet-50-model.caffemodel] [--process=4]

For examples,

python train.py 0 fm-res39 --init_weights=ResNet-50-model.caffemodel

Evaluation

python test.py [gpu_id] [model] [--init_weights=xxx.caffemodel]

python evalCOCO.py [model]