deepmask-pytorch

This repository contains a PyTorch re-implementation for the DeepMask and SharpMask object proposal algorithms.

Requirements and Dependencies

Linux
NVIDIA GPU with compute capability 3.5+
python3
PyTorch 0.4.1

Quick Start

To run pretrained DeepMask models to generate object proposals, follow these steps:

Clone this repository into $DEEPMASK:

git clone https://github.com/foolwood/deepmask-pytorch.git
cd deepmask-pytorch
DEEPMASK=$PWD
export PYTHONPATH=$DEEPMASK:$PYTHONPATH

Download pre-trained DeepMask models:

mkdir -p $DEEPMASK/pretrained/deepmask; cd $DEEPMASK/pretrained/deepmask
wget http://www.robots.ox.ac.uk/~qwang/DeepMask.pth.tar

Run computeProposals.py with a given model and optional target image (specified via the -img option):

# apply to a default sample image (data/testImage.jpg)
cd $DEEPMASK
python tools/computeProposals.py --arch DeepMask --resume $DEEPMASK/pretrained/deepmask/DeepMask.pth.tar --img ./data/test.jpg

Training Your Own Model

To train your own DeepMask models, follow these steps:

Preparation

If you have not done so already, clone this repository into $DEEPMASK:

git clone https://github.com/foolwood/deepmask-pytorch.git
cd deepmask-pytorch
DEEPMASK=$PWD
export PYTHONPATH=$DEEPMASK:$PYTHONPATH

Download and extract the COCO images and annotations:

mkdir -p $DEEPMASK/data/coco; cd $DEEPMASK/data/coco
wget http://images.cocodataset.org/zips/train2017.zip
wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip

unzip ./train2017.zip && unzip ./val2017.zip && unzip ./annotations_trainval2017.zip
cd $DEEPMASK/loader/pycocotools && make

Training

To train DeepMask, launch the train.py script. It contains several options, to list them, simply use the --help flag.

cd $DEEPMASK
CUDA_VISIBLE_DEVICES=0,1,2,3 python tools/train.py --dataset coco -j 20 --freeze_bn
CUDA_VISIBLE_DEVICES=0,1,2,3 python tools/train.py --dataset coco -j 20 --arch SharpMask --freeze_bn

Testing

Test on COCO validation set (5K).

sh scripts/test_recall_coco.sh

Our results

                +-----------------+------------------------+------------------------+
                |                 |      Box Proposals     | Segmentation Proposals |
                +                 +------------------------+------------------------+
                |                 | AR_10 | AR_100 | AR_1K | AR_10 | AR_100 | AR_1K |
                +-----------------+-------+--------+-------+-------+--------+-------+
                | DeepMask(paper) |  18.7 |  34.9  |  46.5 |  14.4 |  25.8  |  33.1 |
                +-----------------+-------+--------+-------+-------+--------+-------+
                | DeepMask(ours)  |  18.3 |  35.9  |  48.4 |  13.6 |  26.0  |  33.5 |
                +-----------------+-------+--------+-------+-------+--------+-------+

Naive Cascade Instance Segmentation (YOLOv3+DeepMask=10FPS~28FPS)

git clone https://github.com/pjreddie/darknet.git
cd darknet
make # Compile with CUDA https://pjreddie.com/darknet/install/
sed -i 's/= data/= .\/darknet\/data/g' cfg/coco.data
sed -i 's/batch=64/batch=1/g' cfg/yolov3.cfg
sed -i 's/subdivisions=16/subdivisions=1/g' cfg/yolov3.cfg
cd $DEEPMASK
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights
python tools/yolo_deepmask.py

Naive Cascade Video Object Segmentation (DaSiamRPN+DeepMask=60FPS)

git clone https://github.com/foolwood/DaSiamRPN.git
mkdir tracker && mv DaSiamRPN/code/* ./tracker/ && rm -rf ./DaSiamRPN
touch tracker/__init__.py
sed -i 's/utils/.utils/g' tracker/run_SiamRPN.py
cd $DEEPMASK
wget www.robots.ox.ac.uk/~qwang/SiamRPNVOT.model
python tools/dasiamrpn_deepmask.py

Citations

Please consider citing this project in your publications if it helps your research. The following is a BibTeX reference. The BibTeX entry requires the url LaTeX package.

@misc{wang2018deepmask,
author = {Wang Qiang},
title = {{deepmask-pytorch}},
year = {2018},
howpublished = {\url{https://github.com/foolwood/deepmask_pytorch}}
}

@inproceedings{DeepMask,
   title = {Learning to Segment Object Candidates},
   author = {Pedro O. Pinheiro and Ronan Collobert and Piotr Dollár},
   booktitle = {NIPS},
   year = {2015}
}
@inproceedings{SharpMask,
   title = {Learning to Refine Object Segments},
   author = {Pedro O. Pinheiro and Tsung-Yi Lin and Ronan Collobert and Piotr Dollár},
   booktitle = {ECCV},
   year = {2016}
}

License

Licensed under an MIT license

foolwood/deepmask-pytorch

foolwood

Reviews

Repository Details