[robotics-worldwide] [software] Code Release: NetVLAD in Python/TensorFlow

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

[robotics-worldwide] [software] Code Release: NetVLAD in Python/TensorFlow

Davide Scaramuzza
Dear colleagues

We are happy to announce a Python/Tensorflow implementation of the place
recognition neural network NetVLAD:
https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_uzh-2Drpg_netvlad-5Ftf-5Fopen&d=DwICaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=0w3solp5fswiyWF2RL6rSs8MCeFamFEPafDTOhgTfYI&m=EhnNcD3wla6tq6G128PQYF6u0f4It9ufdTj-qaL2rUw&s=ONkwWc6zpUcjxB9Q5YrXGeMUj4ljIYTOCBklB33gJPY&e=

NetVLAD [Arandjelovic 2016 CVPR],
https://urldefense.proofpoint.com/v2/url?u=https-3A__www.di.ens.fr_willow_research_netvlad_&d=DwICaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=0w3solp5fswiyWF2RL6rSs8MCeFamFEPafDTOhgTfYI&m=EhnNcD3wla6tq6G128PQYF6u0f4It9ufdTj-qaL2rUw&s=0Vvz6LUKRkJ5yfG2LUGJ9VqaOUxQQUWwdLDDZw8zRZM&e= ] is a place recognition
neural network which takes an image as input and produces a vector as
output. If two images are taken in the same place, the Euclidean between
these vectors is small, otherwise not. Using nearest-neighbors search on
these vectors, the authors have shown excellent place recognition
performance, even under severe appearance changes.

However, the full network has officially so far only been implemented
available Matlab, rendering deployment on non-desktop PCs and robots
tedious.

We are happy to announce a Python/Tensorflow port of the FULL network,
approved by the original authors and available at:
https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_uzh-2Drpg_netvlad-5Ftf-5Fopen&d=DwICaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=0w3solp5fswiyWF2RL6rSs8MCeFamFEPafDTOhgTfYI&m=EhnNcD3wla6tq6G128PQYF6u0f4It9ufdTj-qaL2rUw&s=ONkwWc6zpUcjxB9Q5YrXGeMUj4ljIYTOCBklB33gJPY&e=
The repository contains code which allows plug-and-play python
deployment of the best off-the-shelf model made available by the
authors. We have thoroughly tested that the ported model produces a
similar output to the original Matlab implementation, as well as
excellent place recognition performance on KITTI 00. The repository does
not contain code to train the network, however, it should be easy to
adapt to other models trained in Matlab.

In our own research, we have previously used NetVLAD in [Ye 2017 BMVC]
and [Cieslewski 2018 ICRA], and will continue to use it extensively.

Titus Cieslewski, Relja Arandjelovic, Davide Scaramuzza

--
___________________________________

Prof. Dr. Davide Scaramuzza
Director of the Robotics and Perception Group: https://urldefense.proofpoint.com/v2/url?u=http-3A__rpg.ifi.uzh.ch_people-5Fscaramuzza.html&d=DwICaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=0w3solp5fswiyWF2RL6rSs8MCeFamFEPafDTOhgTfYI&m=EhnNcD3wla6tq6G128PQYF6u0f4It9ufdTj-qaL2rUw&s=yvU7iXNWST_8C6NqL9yInWDo594XIti3klwW3qYbVfo&e=
Inst. of Informatics, University of Zurich,
Inst. of Neuroinformatics, University of Zurich and ETH Zurich
Andreasstrasse 15, AND 2.10, Zurich, Switzerland
Office: +41 44 635 24 09
YouTube Channel: https://urldefense.proofpoint.com/v2/url?u=https-3A__www.youtube.com_ailabRPG_videos&d=DwICaQ&c=clK7kQUTWtAVEOVIgvi0NU5BOUHhpN0H8p7CSfnc_gI&r=0w3solp5fswiyWF2RL6rSs8MCeFamFEPafDTOhgTfYI&m=EhnNcD3wla6tq6G128PQYF6u0f4It9ufdTj-qaL2rUw&s=kzfbA1ohkMEOmgnasW6u6lFSWLkkf_tYr3keYHdIYIA&e=
___________________________________

_______________________________________________
robotics-worldwide mailing list
[hidden email]
http://duerer.usc.edu/mailman/listinfo.cgi/robotics-worldwide