Skip to content
Branch: master
Find file History
Latest commit 3b1399c Aug 23, 2017
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
.gitignore cosmetic. Apr 28, 2015
README.md Update README.md Aug 23, 2017
__init__.py
data_processing.py cosmetic. Apr 16, 2015
dataset.py cosmetics. Apr 28, 2015
download.sh Update download.sh Aug 23, 2017
evaluate.py added distribution distances. May 21, 2015
sift.py h5py loves close()-d stores. Jul 3, 2012

README.md

These are the patches downloadable from [http://www.cs.ubc.ca/~mbrown/patchdata/patchdata.html].

The three datasets liberty, notredame and yosemite:

  • wget -c http://matthewalunbrown.com/patchdata/liberty.zip
  • wget -c http://matthewalunbrown.com/patchdata/notredame.zip
  • wget -c http://matthewalunbrown.com/patchdata/yosemite.zip

Extract all three zip files into liberty/, notredame/, yosemite/ respectively.

Run python dataset.py to build a hdf5 store (default name: patchdata_64x64.h5), the size is about 6GB (saved as uint8!). Use select to build a store useful for training. Look into evaluate.py in order to evaluate pairs on ground truth.

You can’t perform that action at this time.