Benchmark

Listed below are the public results on the three benchmark datasets. The localization results are reported as the percentage of query images which where localized within three given translation and rotation thresholds, for each condition.

Aachen Day-Night dataset

Localization thresholds:

  • Day: (0.25m, 2°) / (0.5m, 5°) / (5m, 10°)
  • Night: (0.5m, 2°) / (1m, 5°) / (5m, 10°)
MethoddaynightLast updated
Visual Localization Using Sparse Semantic 3D Map71.8 / 91.5 / 96.840.8 / 63.3 / 80.6June 2, 2019, 4 a.m.
Hierarchical-Localization NetVLAD+SuperPoint80.5 / 87.4 / 94.242.9 / 62.2 / 76.5June 1, 2019, 9:50 a.m.
Hierarchical-Localization (multi-camera when available)80.5 / 87.4 / 94.242.9 / 62.2 / 76.5June 2, 2019, 6:55 a.m.
DenseVLAD & D2-Net (top-20)80.1 / 88.0 / 93.439.8 / 55.1 / 74.5May 29, 2019, 7:48 a.m.
UR2KID Scape Technologies (paper coming soon, not scene specific, single scale)0.0 / 0.0 / 0.046.9 / 67.3 / 88.8July 3, 2019, 3:39 p.m.
D2-Net - single-scale0.0 / 0.0 / 0.045.9 / 68.4 / 88.8July 16, 2019, 12:21 p.m.
R2D2 10k keypoints0.0 / 0.0 / 0.045.9 / 66.3 / 88.8June 17, 2019, 9:56 a.m.
Asymmetric Hypercolumn Matching47.8 / 72.2 / 91.330.6 / 53.1 / 78.6June 5, 2019, 7:28 p.m.
densecontextdesc10k_upright_mixedmatcher0.0 / 0.0 / 0.046.9 / 65.3 / 87.8July 16, 2019, 11:13 a.m.
D2-Net - single-scale off-the-shelf0.0 / 0.0 / 0.041.8 / 69.4 / 86.7July 16, 2019, 4:57 p.m.
DELF - new model0.0 / 0.0 / 0.039.8 / 61.2 / 85.7June 12, 2019, 3:52 p.m.
NetVLAD+SuperPoint top 10 (baseline)33.9 / 48.8 / 78.525.5 / 45.9 / 80.6June 1, 2019, 8:59 p.m.
CityScaleLocalization52.3 / 80.0 / 94.324.5 / 33.7 / 49.0May 10, 2019, 2:22 p.m.
DELF - old model0.0 / 0.0 / 0.039.8 / 60.2 / 84.7May 21, 2019, 2:52 p.m.
Localizing Visual Landmarks for 2D matching62.4 / 71.8 / 79.924.5 / 35.7 / 44.9April 20, 2019, 5:21 p.m.
saliency_ranking_net (multi-scale)0.0 / 0.0 / 0.044.9 / 59.2 / 77.6April 29, 2019, 2:33 p.m.
Active Search57.3 / 83.7 / 96.619.4 / 30.6 / 43.9March 13, 2019, 9 a.m.
contextdesc10k_upright0.0 / 0.0 / 0.037.8 / 56.1 / 80.6July 16, 2019, 11:17 a.m.
DGCNCCC22.9 / 49.8 / 84.719.4 / 37.8 / 68.4May 30, 2019, 10:17 a.m.
HesAffNet-HardNet20.0 / 0.0 / 0.037.8 / 54.1 / 75.5May 31, 2019, 10:35 p.m.
Upright RootSIFT (Feature Challenge Baseline)0.0 / 0.0 / 0.033.7 / 52.0 / 65.3May 2, 2019, 8:57 a.m.
ELF0.0 / 0.0 / 0.013.3 / 21.4 / 30.6May 3, 2019, 8:36 p.m.
DenseVLAD0.0 / 0.1 / 22.80.0 / 2.0 / 14.3April 11, 2019, 5:43 p.m.
NetVLAD0.0 / 0.2 / 18.90.0 / 2.0 / 12.2May 31, 2019, 11:16 p.m.
Localizing Visual Landmarks for Place Recognition0.0 / 0.2 / 20.80.0 / 1.0 / 10.2June 2, 2019, 10:06 p.m.
DGCNCCC0.0 / 0.1 / 9.50.0 / 0.0 / 8.2March 17, 2019, 11:42 a.m.
NetVlad_NN_Pose0.0 / 0.1 / 7.00.0 / 0.0 / 4.1March 17, 2019, 11:46 a.m.
joint image segemation and depth to posenet0.0 / 0.1 / 24.90.0 / 0.0 / 0.0May 31, 2019, 11:45 a.m.
FAB-MAP0.0 / 0.0 / 4.60.0 / 0.0 / 0.0March 13, 2019, 11:54 a.m.

CMU Seasons dataset

Localization thresholds:

  • All conditions: (0.25m, 2°) / (0.5m, 5°) / (5m, 10°)
MethodurbansuburbanparkLast updated
Hierarchical-Localization NetVLAD+SuperPoint91.7 / 94.6 / 97.774.5 / 81.5 / 91.354.3 / 62.5 / 79.0June 1, 2019, 9:50 a.m.
LocalSfM (Upright RootSIFT) (has partial access to ground truth)72.8 / 74.1 / 76.155.2 / 57.7 / 61.341.8 / 44.5 / 48.7March 13, 2019, 11:42 a.m.
Semantic Match Consistency75.0 / 82.1 / 87.844.0 / 53.6 / 63.730.0 / 37.9 / 48.2March 18, 2019, 11:47 p.m.
EffecientPairWise2D-3DMatching64.1 / 67.8 / 71.530.2 / 35.5 / 40.923.8 / 28.1 / 34.0March 20, 2019, 9:52 a.m.
DenseVLAD22.2 / 48.6 / 92.89.8 / 26.6 / 85.210.3 / 27.1 / 77.0April 15, 2019, 11:03 a.m.
Active Search55.2 / 60.3 / 65.120.7 / 25.9 / 29.912.7 / 16.3 / 20.8March 13, 2019, 9:04 a.m.
NetVLAD17.4 / 40.3 / 93.27.6 / 21.0 / 80.55.6 / 15.7 / 65.8April 11, 2019, 9:16 p.m.
CityScaleLocalization36.7 / 42.0 / 53.18.6 / 11.7 / 21.17.0 / 9.6 / 17.0March 13, 2019, 11:59 a.m.
FAB-MAP2.7 / 6.4 / 27.30.5 / 1.5 / 13.60.8 / 1.7 / 11.5March 13, 2019, 11:49 a.m.

Extended CMU Seasons dataset

Localization thresholds:

  • All conditions: (0.25m, 2°) / (0.5m, 5°) / (5m, 10°)
MethodurbansuburbanparkLast updated
Hierarchical-Localization (multi-camera when available)91.6 / 96.4 / 99.184.7 / 91.5 / 98.669.3 / 77.8 / 90.5June 2, 2019, 6:55 a.m.
Visual Localization Using Sparse Semantic 3D Map88.8 / 93.6 / 96.378.0 / 83.8 / 89.263.6 / 70.3 / 77.3June 2, 2019, 4 a.m.
Hierarchical-Localization NetVLAD+SuperPoint89.5 / 94.2 / 97.976.5 / 82.7 / 92.757.4 / 64.4 / 80.4June 1, 2019, 9:50 a.m.
Asymmetric Hypercolumn Matching65.7 / 82.7 / 91.066.5 / 82.6 / 92.954.3 / 71.6 / 84.1June 5, 2019, 7:24 p.m.
Localizing Visual Landmarks for 2D matching84.3 / 89.3 / 93.068.0 / 75.1 / 84.442.4 / 51.4 / 69.7April 25, 2019, 8:12 a.m.
CityScaleLocalization71.2 / 74.6 / 78.757.8 / 61.7 / 67.534.5 / 37.0 / 42.2March 8, 2019, 8:20 p.m.
Localizing Visual Landmarks for Place Recognition17.3 / 42.5 / 89.05.8 / 19.4 / 76.16.6 / 23.1 / 73.0June 2, 2019, 10:07 p.m.
DGCNCCC17.1 / 41.5 / 89.18.9 / 26.8 / 77.14.8 / 16.2 / 63.3June 2, 2019, 1:27 p.m.
Netvlad_localization16.4 / 38.5 / 78.98.8 / 27.1 / 77.04.7 / 16.0 / 63.2June 25, 2019, 9:44 p.m.
DenseVLAD14.7 / 36.3 / 83.95.3 / 18.7 / 73.95.2 / 19.1 / 62.0March 12, 2019, 3:21 p.m.
NetVLAD12.2 / 31.5 / 89.83.7 / 13.9 / 74.72.6 / 10.4 / 55.9June 2, 2019, 9:12 a.m.
joint image segemation and depth to posenet0.3 / 1.5 / 31.90.1 / 0.3 / 12.50.0 / 0.2 / 9.0May 31, 2019, 11:46 a.m.

RobotCar Seasons dataset

Localization thresholds:

  • All conditions: (0.25m, 2°) / (0.5m, 5°) / (5m, 10°)
Methodday allnight allLast updated
Visual Localization Using Sparse Semantic 3D Map54.5 / 81.6 / 96.712.3 / 28.5 / 46.5June 2, 2019, 4 a.m.
Hierarchical-Localization (multi-camera when available)53.8 / 80.4 / 96.011.2 / 27.7 / 49.1June 2, 2019, 6:54 a.m.
Semantic Match Consistency50.3 / 79.3 / 95.27.1 / 22.4 / 45.3March 18, 2019, 11:47 p.m.
Active Search on sequences of camera triplets (uses ground truth relative poses)46.6 / 80.1 / 97.05.8 / 21.0 / 43.1March 13, 2019, 12:03 p.m.
Hierarchical-Localization NetVLAD+SuperPoint53.1 / 79.1 / 95.57.2 / 17.4 / 34.4June 1, 2019, 9:50 a.m.
Asymmetric Hypercolumn Matching45.7 / 78.0 / 95.122.3 / 61.8 / 94.5June 5, 2019, 7:25 p.m.
Localizing Visual Landmarks for 2D matching51.1 / 77.7 / 92.313.8 / 30.3 / 57.7May 16, 2019, 2:09 a.m.
EffecientPairWise2D-3DMatching42.6 / 78.2 / 95.78.5 / 14.1 / 20.6March 20, 2019, 9:52 a.m.
Scalable and Accurate Image-based Localization48.0 / 78.0 / 94.23.4 / 9.5 / 17.0May 9, 2019, 9:02 a.m.
Active Seach on Camera Triplets45.5 / 77.0 / 94.72.7 / 6.9 / 12.1March 13, 2019, 9:31 a.m.
CityScaleLocalization45.3 / 73.5 / 90.10.6 / 2.6 / 7.2March 13, 2019, 11:59 a.m.
Active Search35.6 / 67.9 / 90.40.9 / 2.1 / 4.3March 13, 2019, 9:04 a.m.
DenseVLAD (single-scale, top-1 interpolation)9.8 / 32.8 / 90.71.8 / 6.6 / 25.6May 29, 2019, 3:27 p.m.
ToDayGAN + DenseVLAD7.6 / 31.2 / 91.22.2 / 10.8 / 50.5May 27, 2019, 8:05 p.m.
DenseVLAD (single-scale)8.1 / 31.8 / 90.71.3 / 5.7 / 25.5May 29, 2019, 1 p.m.
DenseVLAD7.6 / 31.2 / 91.21.0 / 4.4 / 22.7April 11, 2019, 5:44 p.m.
Localizing Visual Landmarks for Place Recognition7.9 / 30.0 / 85.94.1 / 15.7 / 59.1June 2, 2019, 10:07 p.m.
NetVLAD6.4 / 26.3 / 90.90.3 / 2.3 / 15.9June 2, 2019, 9:12 a.m.
COCL_HR12.7 / 60.6 / 75.40.5 / 1.6 / 2.2May 30, 2019, 9:20 a.m.
DGCNCCC7.2 / 28.3 / 87.60.0 / 1.1 / 13.4May 21, 2019, 7:38 a.m.
FAB-MAP2.7 / 11.8 / 37.30.0 / 0.0 / 0.0March 13, 2019, 11:50 a.m.
joint image segemation and depth to posenet0.1 / 0.8 / 31.00.0 / 0.0 / 0.0May 31, 2019, 11:46 a.m.

InLoc dataset

Localization thresholds:

  • All conditions: (0.25m, 10°) / (0.5m, 10°) / (1m, 10°)
Methodduc1duc2Last updated
Visual Localization Using Sparse Semantic 3D Map41.4 / 59.1 / 71.238.2 / 49.6 / 58.0June 2, 2019, 4 a.m.
InLoc40.9 / 58.1 / 70.235.9 / 54.2 / 69.5April 30, 2019, 1:52 p.m.
DisLoc+SparsePE23.7 / 34.3 / 49.516.8 / 24.4 / 32.8May 30, 2019, 9:57 a.m.
DenseVLAD0.0 / 1.5 / 5.10.0 / 0.8 / 2.3April 30, 2019, 3:02 p.m.

SILDa Weather and Time of Day dataset

Localization thresholds:

  • All conditions: (0.25m, 2°) / (0.5m, 5°) / (5m, 10°)
MethodeveningsnownightLast updated
Visual Localization Using Sparse Semantic 3D Map29.4 / 66.5 / 93.42.2 / 11.1 / 57.420.4 / 41.8 / 69.0June 2, 2019, 4 a.m.
DenseVLAD (top-20) & D2-Net (single scale)24.2 / 58.0 / 76.01.0 / 6.3 / 48.123.2 / 50.2 / 79.4June 8, 2019, 9:18 p.m.
DenseVLAD (top-20) + Colmap Image Registrator28.2 / 56.2 / 71.23.1 / 9.4 / 36.816.3 / 33.4 / 50.7April 30, 2019, 12:24 p.m.
Localizing Visual Landmarks for Place Recognition0.3 / 7.1 / 41.00.0 / 0.0 / 0.90.0 / 2.7 / 56.9June 2, 2019, 10:06 p.m.
DenseVLAD0.2 / 6.2 / 42.10.0 / 0.0 / 0.70.0 / 2.6 / 53.9April 18, 2019, 12:59 p.m.
DGCNCCC0.0 / 0.0 / 0.00.0 / 0.0 / 0.00.0 / 0.0 / 0.0May 19, 2019, 1:16 p.m.