Artcs1 / Keypoints

Keypoints

This is the official repository of the OMNICV-CVPR2022 workshop named "Pose Estimation for Two-View Panoramas: a Comparative Analysis"

Visit our paper:

IEEE link

CVF access

Installation

1. Create a conda enviorenment

conda create --name py37-keypoints python=3.7
conda activate py37-keypoints

2. Install the Spherical Package from Meder github

3. Git clone the project and enter the folder

git clone https://github.com/Artcs1/Keypoints.git
cd Keypoints

4. Install the requirements

pip install -r requirements.txt

5. Install liegroups package in utils directory

6. Compile the SPHORB package

cd SPHORB-master
conda create --name SPHORB python=3.7
conda activate SPHORB
conda install -c conda-forge opencv==3.4.2
mkdir build1
cd build1
cmake ..
make
conda deactivate

7. Compile the five-point algorithm package

cd fivepoint
conda create --name FP python=3.7
conda activate FP
conda install -c conda-forge opencv==4.5.5
mkdir build
cd build
cmake ..
make
conda deactivate

Usage

1. Activate the conda enviorenment

conda activate py37-keypoints

2. Run extrack_keypoints.py file

python3 extract_keypoints.py --datas [Urban1|Urban2|Urban3|Urban4|Room|Classroom|Realistic|Interior1|Interior2] --descriptors [sift|tsift|orb|torb|spoint|tspoint|sphorb]

3. Example for Outdoor dataset

python3 extract_keypoints.py --datas Urban1 Urban2 Urban3 Urban4 --descriptors  orb torb sift tsift spoint tspoint sphorb

4. Example for Indoor dataset

python3 extract_keypoints.py --datas Room Classroom Realistic Interior1 Interior2 --descriptors  orb torb sift tsift spoint tspoint sphorb

Visual example

python3 show_keypoints.py --descriptor sift

Benchmark and results

100 images per dataset

Downloand link

1000 images per dataset

Downloand link

Cite our work

@InProceedings{Murrugarra-Llerena_2022_CVPR,
    author    = {Murrugarra-Llerena, Jeffri and da Silveira, Thiago L. T. and Jung, Claudio R.},
    title     = {Pose Estimation for Two-View Panoramas Based on Keypoint Matching: A Comparative Study and Critical Analysis},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2022},
    pages     = {5202-5211}
}