TILDE: A Temporally Invariant Learned DEtector


We introduce a learning-based approach to detect repeatable keypoints under drastic imaging changes of weather and lighting conditions to which state-of-the-art keypoint detectors are surprisingly sensitive. We first identify good keypoint candidates in multiple training images taken from the same viewpoint. We then train a regressor to predict a score map whose maxima are those points so that they can be found by simple non-maximum suppression. As there are no standard datasets to test the influence of these kinds of changes, we created our own, which we will make publicly available. We will show that our method significantly outperforms the state-of-the-art methods in such challenging conditions, while still achieving state-of-the-art performance on the untrained standard Oxford dataset.


TILDE: A Temporally Invariant Learned DEtector

Y. Verdie; K. M. Yi; P. Fua; V. Lepetit 

2015. Computer Vision and Pattern Recognition (CVPR), Boston, Massachusetts, USA, p. 5279-5288. DOI : 10.1109/CVPR.2015.7299165.

Teaser Video

This video teaser shows additional results of our methods on sequences which were not shown on the paper such as Mexico and Chamonix sequences. A cross is displayed when not enough matches are provided to compute an homography. To avoid the effect of the descriptors, we use a nearest neighbor match for creating the pairs.

This is another video showing the actual TILDE keypoints detected on the Chamonix sequence. Note that our points are quite reliable compared to SIFT. Best 100 keypoints for both methods are shown.

Example application

This video illustrates a potential applications directly using our method. A video is created from a cellphone and matched to a panoramic image generated beforehand. As there is a large temporal gap between the two sequences, it is difficult for traditional methods to find correspondance between the two. We use the same setting for both methods and ‘Opponent SIFT’ for descriptor.

Supplementary material

Click the following link for the supplementary appendix for implementation details and mathematical derivations.