Skip to content

Commit

Permalink
Typo.
Browse files Browse the repository at this point in the history
  • Loading branch information
@etrulls
etrulls committed Apr 16, 2018
1 parent 9535653 commit 7168ea6
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions readme.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -119,8 +119,8 @@ submission](https://github.com/cvlab-epfl/LIFT)), so there are some changes in
the architecture and training procedure. Performance should be about on par.

The files can be downloaded here:
* [Models without rotation augmentation](http://webhome.cs.uvic.ca/~kyi/files/2018/tflift/releasea-aug.tar.gz) (run with --use_batch_norm=False --mean_std_type=dataset)
* [Models with rotation augmentation](http://webhome.cs.uvic.ca/~kyi/files/2018/tflift/releasea-no-aug.tar.gz) (run with --use_batch_norm=False --mean_std_type=hardcoded)
* [Models without rotation augmentation](http://webhome.cs.uvic.ca/~kyi/files/2018/tflift/release-no-aug.tar.gz) (run with --use_batch_norm=False --mean_std_type=dataset)
* [Models with rotation augmentation](http://webhome.cs.uvic.ca/~kyi/files/2018/tflift/release-aug.tar.gz) (run with --use_batch_norm=False --mean_std_type=hardcoded)

The models trained without rotation augmentation perform better on matching
problems where the images are generally upright. For data with random
Expand Down

0 comments on commit 7168ea6

Please sign in to comment.