Python demo for our CVPR'2019 paper
This code was tested with Python 3.8 and mosek 9.2.29. The easiest way to install mosek is:
conda install -c mosek mosek
Please, execute 'demo.py' to view our demo.
This demo does the following using the concepts introduced in our paper:
- Find landmarks in a subsequence of the Oxford Robotcar run from 2015-10-29 12:18:17
- Match a short query sequence from 2014-11-18 13:20:12
The precalculated feature distances in this demo are based on features extracted with a VGG-16 + NetVLAD + whitening network. We use the Off-the-shelf on Pitts30k model available on the NetVLAD project page in combination with this NetVLAD TensorFlow implementation.
If you do not have mosek installed, you can have a look at the saved figures in the results folder instead.
The produced outputs are:
- Scatter plot of original reference and query sequences
- Topology of reference sequence used for finding landmarks with network flow
- Selected landmarks
- Accuracy vs. distance plot of the final matching