Skip to content

Files

Latest commit

author
Samik Some
Update README
Sep 6, 2020
da60b7f · Sep 6, 2020

History

History
127 lines (107 loc) · 3.53 KB

README.md

File metadata and controls

127 lines (107 loc) · 3.53 KB

Faster-RCNN with DPP

This code is part of the paper: Determinantal Point Process as an alternative to NMS published at BMVC 2020.

The code is forked from and based on A Faster Pytorch Implementation of Faster R-CNN and as such can be used in a similar fashion.

Prerequisites

  • Python 3.7.5
  • PyTorch 1.5.0
  • Boost 1.71.0
  • Eigen 3.3.7
  • CUDA 10.0

Installation

  • Clone the repository and checkout the relevant branch.
git clone https://github.com/samiksome/faster-rcnn.pytorch
cd faster-rcnn.pytorch
git checkout dpp
  • Install requirements
pip install -r requirements.txt
  • Build
cd lib
python setup.py build develop
cd model
make
  • Replace pycocotools
cd ../../..
git clone "https://github.com/cocodataset/cocoapi"
cd cocoapi/PythonAPI

make all

cd ../../faster-rcnn.pytorch/lib
rm -rf pycocotools
cp -r ../../cocoapi/PythonAPI/pycocotools ./
cd ..

Datasets and Pretrained models

  • Get datasets (PASCAL trainval and COCO train datasets can be skipped if training is not needed)
mkdir data
cd data

wget "http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar"
wget "http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar"
wget "http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCdevkit_08-Jun-2007.tar"
tar -xvf VOCtrainval_06-Nov-2007.tar
tar -xvf VOCtest_06-Nov-2007.tar
tar -xvf VOCdevkit_08-Jun-2007.tar
mv VOCdevkit VOCdevkit2007

mkdir coco
cd coco

wget "http://images.cocodataset.org/zips/train2014.zip"
wget "http://images.cocodataset.org/zips/val2014.zip"
wget "http://images.cocodataset.org/annotations/annotations_trainval2014.zip"
wget "https://dl.dropboxusercontent.com/s/o43o90bna78omob/instances_minival2014.json.zip"
wget "https://dl.dropboxusercontent.com/s/s3tw5zcg7395368/instances_valminusminival2014.json.zip"
unzip train2014.zip
unzip val2014.zip
unzip annotations_trainval2014.zip
unzip instances_minival2014.json.zip
unzip instances_valminusminival2014.json.zip
mkdir images
mv train2014 images/
mv val2014 images/
mv instances_minival2014.json annotations/
mv instances_valminusminival2014.json annotations/

cd ../..
  • Download trained models (ours)
wget "https://www.dropbox.com/s/dnxsdhkhcj9jvn0/models.zip"
unzip models.zip

Running the code

  • Test on PASCAL VOC (with our trained models) using NMS
python test_net.py --dataset pascal_voc \
                   --net vgg16 \
                   --checksession 1 --checkepoch 6 --checkpoint 10021 \
                   --cuda
  • Test on PASCAL VOC (with our trained models) using DPP
python test_net.py --dataset pascal_voc \
                   --net vgg16 \
                   --checksession 1 --checkepoch 6 --checkpoint 10021 \
                   --use_dpp --dpp_alpha 5 \
                   --cuda

Replace --dataset pascal_voc with --dataset coco and --checkpoint 10021 with --checkpoint 58632 to test on COCO dataset.

By default the maximum number of windows selected (k) is 300. To use a different k change line 196 in faster-rcnn.pytorch/lib/model/utils/config.py to required value.

__C.TEST.RPN_POST_NMS_TOP_N = 300

Citation

Please cite the following paper if you use this code:

@misc{some2020determinantal,
    title={Determinantal Point Process as an alternative to NMS},
    author={Samik Some and Mithun Das Gupta and Vinay P. Namboodiri},
    year={2020},
    eprint={2008.11451},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}