List your company
ResearchFormerly WizWand

Image Retrieval on Oxford5k (test)

91.9mAPGeM+alphaQE
3.526.4549.472.35Nov 18, 2015Apr 27, 2016Oct 6, 2016Mar 17, 2017Aug 25, 2017Feb 3, 2018Jul 15, 2018
Updated 8mo ago

Evaluation results

MethodLinks
GeM+alphaQENet=VGG, F-tuned=yes, Dim=512, Representation=multi-scale, Whitening=Lw2017.1191.9
HesAffNet-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1191.1
GeM+alphaQENet=Res, F-tuned=yes, Dim=2048, Representation=multi-scale, Whitening=Lw2017.1191
R-MAC+QENet=Res, F-tuned=yes, Dim=20482017.1190.6
R-MAC+QENet=VGG, F-tuned=yes, Dim=5122017.1189.1
HesAffNet-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1189
HQENet=n/a, F-tuned=n/a, Dim=n/a2017.1188
HQE(65k)Search Strategy=Query Expansion, Query Cropping=Crop2016.0488
HesAffNet-RootSIFTRetrieval Pipeline=+SV+QE(8)2017.1188
GeMNet=VGG, F-tuned=yes, Dim=512, Representation=multi-scale, Whitening=Lw2017.1187.9
GeMNet=Res, F-tuned=yes, Dim=2048, Representation=multi-scale, Whitening=Lw2017.1187.8
HesAffNetLess-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1187.3
HesAffNet-RootSIFTRetrieval Pipeline=+SV+QE(15)2017.1186.5
R-MACNet=Res, F-tuned=yes, Dim=20482017.1186.1
HesAffNetLess-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1186.1
HesAff-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1185.1
MAC+R+QEDimensionality=512, Backbone=VGG, Training Protocol=Fine-tuned, Search Strategy=Re-ranking and Query Expansion, Query Cropping=Crop2016.0485
Mikulik et al. (2013)2015.1184.9
BoW-fVocab+R+QENet=n/a, F-tuned=n/a, Dim=n/a2017.1184.9
HesAff-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1184.5
R-MACNet=VGG, F-tuned=yes, Dim=5122017.1183.1
Chum et al. (2011)2015.1182.7
BoW+R+QENet=n/a, F-tuned=n/a, Dim=n/a2017.1182.7
Fisher VectorNet=VGG, F-tuned=yes, Dim=5122017.1181.5
Danfeng et al. (2011)2015.1181.4
Tolias et al. (2015)2015.1180.4
HesAff-RootSIFTRetrieval Pipeline=+SV+QE(8)2017.1180.1
MACDimensionality=512, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop X2016.0480
MACDimensionality=512, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop2016.0479.7
MAC-Hard(V)Backbone=VGG162016.1179.7
R-MAC+QENet=Res, F-tuned=no, Dim=20482017.1178.9
BOW-CNN+R+QENet=VGG, F-tuned=no, Dim=n/a2017.1178.8
HesAff-RootSIFTRetrieval Pipeline=+SV+QE(15)2017.1178.4
Tao et al. (2014)2015.1177.8
HesAffNet-HardNet++Retrieval Pipeline=+SV2017.1177.8
HesAffNet-TFeat-M*Retrieval Pipeline=+SV+QE(8)2017.1177.5
R-MAC+AML+QE2015.1177.3
R-MAC+R+QENet=VGG, F-tuned=no, Dim=5122017.1177.3
R-MAC-HardBackbone=VGG162016.1177
MACDimensionality=128, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop X2016.0476.8
R-MACDimensionality=128, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop X2016.0476.7
Proposed Siamese NetworkBackbone=VGG162016.1176.4
MACDimensionality=128, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop2016.0475.8
Shen et al. (2014)2015.1175.2
HesAffNet-TFeat-M*Retrieval Pipeline=+SV+QE(15)2017.1175.2
CroW+QENet=VGG, F-tuned=no, Dim=5122017.1174.9
BoW-CNNNet=VGG, F-tuned=no, Dim=n/a2017.1173.9
HesAff-TFeat-M*Retrieval Pipeline=+SV+QE(8)2017.1173.8
HesAffNetLess-HardNet++Retrieval Pipeline=+SV2017.1173.3
HesAffNet-RootSIFTRetrieval Pipeline=+SV2017.1172.8
R-MACDimensionality=128, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop2016.0472.5
ImageNet labelsBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.0772.4
HesAff-TFeat-M*Retrieval Pipeline=+SV+QE(15)2017.1172.2
NetVLADNet=VGG, F-tuned=yes, Dim=40962017.1171.6
CroWNet=VGG, F-tuned=no, Dim=5122017.1170.8
Finetuned-BaselineBackbone=VGG162016.1169.8
HesAff-HardNet++Retrieval Pipeline=+SV2017.1169.6
R-MACNet=Res, F-tuned=no, Dim=20482017.1169.4
HesAffNet-HardNet++Retrieval Pipeline=BoW2017.1168.3
CroWBackbone=VGG162016.1168.2
SPOCNet=VGG, F-tuned=no, Dim=5122017.1168.1
NetVLADNet=VGG, F-tuned=yes, Dim=5122017.1167.6
R-MACD=5122015.1166.9
R-MACNet=VGG, F-tuned=no, Dim=5122017.1166.9
R-MACBackbone=VGG162016.1166.9
NetVLADNet=VGG, F-tuned=no, Dim=40962017.1166.6
Proposed Siamese NetworkBackbone=CaffeNet2016.1166.2
HesAffNetLess-HardNet++Retrieval Pipeline=BoW2017.1164.3
HesAff-RootSIFTRetrieval Pipeline=+SV2017.1163
R-MAC-HardBackbone=CaffeNet2016.1162.5
MAC-Hard(V)Backbone=CaffeNet2016.1162.2
HesAffNet-RootSIFTRetrieval Pipeline=BoW2017.1161.6
DeepClusterBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.0761
HesAff-HardNet++Retrieval Pipeline=BoW2017.1160.8
Finetuned-BaselineBackbone=CaffeNet2016.1160.2
HesAffNet-TFeat-M*Retrieval Pipeline=+SV2017.1157.3
MACNet=VGG, F-tuned=no, Dim=5122017.1156.4
R-MACD=2562015.1156.1
R-MACBackbone=CaffeNet2016.1156.1
Jégou & Zisserman (2014)D=10242015.1156
Babenko et al. (2014)D=1282015.1155.7
Neural codesDimensionality=128, Backbone=AlexNet, Training Protocol=Fine-tuned, Query Cropping=Crop2016.0455.7
MACDimensionality=128, Backbone=VGG, Training Protocol=Off-the-shelf, Query Cropping=Crop X2016.0455.7
Neural codesBackbone=CaffeNet2016.1155.7
HesAff-TFeat-M*Retrieval Pipeline=+SV2017.1155.6
HesAff-RootSIFTRetrieval Pipeline=BoW2017.1155.1
MACDimensionality=128, Backbone=VGG, Training Protocol=Off-the-shelf, Query Cropping=Crop2016.0453.5
Razavian et al. (2014b)D=2562015.1153.3
Babenko & Lempitsky (2015)D=2562015.1153.1
mVoc/BoWDimensionality=128, Query Cropping=Crop2016.0448.8
mVoc/BoWBackbone=CaffeNet2016.1148.8
HesAff-TFeat-M*Retrieval Pipeline=BoW2017.1146.7
HesAffNet-TFeat-M*Retrieval Pipeline=BoW2017.1145.5
Jégou & Zisserman (2014)D=1282015.1143.3
Wang et al.Backbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=false2018.0742.3
Doersch et al.Backbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=false2018.0735.4
RandomBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.076.9