pySLAM：多深度学习特征集成SLAM

简介

今天推荐一个有趣的项目pySLAM，该用项目用python实现SLAM、VO、关键帧、BA、特征匹配等功能。

最重要的是该项目集成了多种近几年主流的深度学习特征点+描述子，该项目可以比较轻松的利用现有的深度学习特征测试SLAM/VO的性能。

感兴趣的同学可以尝试下这个项目，关注本号后台回复pyslam查看源代码，另外博客地址[1]

目前已支持下述特征检测器：

• FAST[2]
• Good features to track[3]
• ORB[4]
• ORB2[5] (improvements of ORB-SLAM2 to ORB detector)
• SIFT[6]
• SURF[7]
• KAZE[8]
• AKAZE[9]
• BRISK[10]
• AGAST[11]
• MSER[12]
• StarDector/CenSurE[13]
• Harris-Laplace[14]
• SuperPoint[15]
• D2-Net[16]
• DELF[17]
• Contextdesc[18]
• LFNet[19]
• R2D2[20]
• Key.Net[21]

已支持下述特征描述子：

• ORB[22]
• SIFT[23]
• ROOT SIFT[24]
• SURF[25]
• AKAZE[26]
• BRISK[27]
• FREAK[28]
• SuperPoint[29]
• Tfeat[30]
• BOOST_DESC[31]
• DAISY[32]
• LATCH[33]
• LUCID[34]
• VGG[35]
• Hardnet[36]
• GeoDesc[37]
• SOSNet[38]
• L2Net[39]
• Log-polar descriptor[40]
• D2-Net[41]
• DELF[42]
• Contextdesc[43]
• LFNet[44]
• R2D2[45]

关于作者

本项目的作者是Luigi Freda，于2007年在罗马大学获得计算机系统工程博士学位，目前是一名自由职业者，目前从事计算机视觉、机器人和机器学习。

参考资料

[1]

博客地址: https://www.luigifreda.com/2020/05/07/my-new-pyslam-v2-is-out/

[2]

FAST: https://www.edwardrosten.com/work/fast.html

[3]

Good features to track: https://ieeexplore.ieee.org/document/323794

[4]

ORB: http://www.willowgarage.com/sites/default/files/orb_final.pdf

[5]

ORB2: https://github.com/raulmur/ORB_SLAM2

[6]

SIFT: https://www.cs.ubc.ca/~lowe/papers/iccv99.pdf

[7]

SURF: http://people.ee.ethz.ch/~surf/eccv06.pdf

[8]

KAZE: https://www.doc.ic.ac.uk/~ajd/Publications/alcantarilla_etal_eccv2012.pdf

[9]

AKAZE: http://www.bmva.org/bmvc/2013/Papers/paper0013/paper0013.pdf

[10]

BRISK: http://www.margaritachli.com/papers/ICCV2011paper.pdf

[11]

AGAST: http://www.i6.in.tum.de/Main/ResearchAgast

[12]

MSER: http://cmp.felk.cvut.cz/~matas/papers/matas-bmvc02.pdf

[13]

StarDector/CenSurE: https://link.springer.com/content/pdf/10.1007%2F978-3-540-88693-8_8.pdf

[14]

Harris-Laplace: https://www.robots.ox.ac.uk/~vgg/research/affine/det_eval_files/mikolajczyk_ijcv2004.pdf

[15]

SuperPoint: https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork

[16]

D2-Net: https://github.com/mihaidusmanu/d2-net

[17]

DELF: https://github.com/tensorflow/models/blob/master/research/delf/INSTALL_INSTRUCTIONS.md

[18]

Contextdesc: https://github.com/lzx551402/contextdesc

[19]

LFNet: https://github.com/vcg-uvic/lf-net-release

[20]

R2D2: https://github.com/naver/r2d2

[21]

Key.Net: https://github.com/axelBarroso/Key.Net

[22]

ORB: http://www.willowgarage.com/sites/default/files/orb_final.pdf

[23]

SIFT: https://www.cs.ubc.ca/~lowe/papers/iccv99.pdf

[24]

ROOT SIFT: https://www.robots.ox.ac.uk/~vgg/publications/2012/Arandjelovic12/arandjelovic12.pdf

[25]

SURF: http://people.ee.ethz.ch/~surf/eccv06.pdf

[26]

AKAZE: http://www.bmva.org/bmvc/2013/Papers/paper0013/paper0013.pdf

[27]

BRISK: http://www.margaritachli.com/papers/ICCV2011paper.pdf

[28]

FREAK: https://www.researchgate.net/publication/258848394_FREAK_Fast_retina_keypoint

[29]

SuperPoint: https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork

[30]

Tfeat: https://github.com/vbalnt/tfeat

[31]

BOOST_DESC: https://www.labri.fr/perso/vlepetit/pubs/trzcinski_pami15.pdf

[32]

DAISY: https://ieeexplore.ieee.org/document/4815264

[33]

LATCH: https://arxiv.org/abs/1501.03719

[34]

LUCID: https://pdfs.semanticscholar.org/85bd/560cdcbd4f3c24a43678284f485eb2d712d7.pdf

[35]

VGG: https://www.robots.ox.ac.uk/~vedaldi/assets/pubs/simonyan14learning.pdf

[36]

Hardnet: https://github.com/DagnyT/hardnet.git

[37]

GeoDesc: https://github.com/lzx551402/geodesc.git

[38]

SOSNet: https://github.com/yuruntian/SOSNet.git

[39]

L2Net: https://github.com/yuruntian/L2-Net

[40]

Log-polar descriptor: https://github.com/DagnyT/hardnet_ptn.git

[41]

D2-Net: https://github.com/mihaidusmanu/d2-net

[42]

DELF: https://github.com/tensorflow/models/blob/master/research/delf/INSTALL_INSTRUCTIONS.md

[43]

Contextdesc: https://github.com/lzx551402/contextdesc

[44]

LFNet: https://github.com/vcg-uvic/lf-net-release

[45]

R2D2: https://github.com/naver/r2d2

本文仅做学术分享，如有侵权，请联系删文。