Submission details

Method nameDenseVLAD & D2-Net
Datasetextended-cmu
Submission dateSept. 20, 2019, 5:28 p.m.
Last modifiedFeb. 25, 2020, 9:10 p.m.
Publication urlhttp://openaccess.thecvf.com/content_CVPR_2019/papers/Dusmanu_D2-Net_A_Trainable_CNN_for_Joint_Description_and_Detection_of_CVPR_2019_paper.pdf
Code urlhttps://github.com/mihaidusmanu/d2-net
Other infoUses DenseVLAD (single-scale) to identify the top-20 closest database images, then uses single-scale D2-Net features for feature matching and COLMAP's image_registrator for localization. Only "rear" images are considered.

Detailed results for all conditions

Methodurbansuburbanparkovercastsunnyfoliagemixed foliageno foliagelow suncloudysnow
DenseVLAD & D2-Net94.0 / 97.7 / 99.193.0 / 95.7 / 98.389.2 / 93.2 / 95.092.5 / 95.6 / 97.586.2 / 91.8 / 95.288.0 / 92.8 / 95.994.3 / 96.9 / 98.498.0 / 99.4 / 99.895.1 / 97.5 / 98.795.5 / 97.5 / 98.997.2 / 98.9 / 99.6