List your company
ResearchFormerly WizWand

Image Retrieval on Paris6k (test)

97.8mAPProposed diffusion w/ late fusion
18.96839.43459.980.366Nov 18, 2015May 20, 2016Nov 20, 2016May 23, 2017Nov 23, 2017May 26, 2018Nov 27, 2018
Updated 8mo ago

Evaluation results

MethodLinks
Proposed diffusion w/ late fusionFeature=R-MAC (ResNet), Global=true, Regional=true2018.1197.8
Proposed diffusionFeature=R-MAC (ResNet), Global=false, Regional=true2018.1197.6
Proposed diffusionFeature=R-MAC (ResNet), Global=true, Regional=false2018.1197.1
Iscen's diffusionFeature=R-MAC (ResNet), Global=true, Regional=true2018.1196.9
Iscen's diffusionFeature=R-MAC (ResNet), Global=true, Regional=false2018.1196.5
Iscen's diffusionFeature=R-MAC (VGG), Global=true, Regional=true2018.1196.5
Proposed diffusion w/ late fusionFeature=R-MAC (VGG), Global=true, Regional=true2018.1196.1
R-MAC+QENet=Res, F-tuned=yes, Dim=20482017.1196
GeM+alphaQENet=Res, F-tuned=yes, Dim=20482017.1195.5
R-match + AQEFeature=R-MAC (ResNet), Global=false, Regional=true2018.1195.5
k-NN + AQEFeature=R-MAC (ResNet), Global=true, Regional=false2018.1195.3
R-matchFeature=R-MAC (ResNet), Global=false, Regional=true2018.1194.9
Proposed diffusionFeature=R-MAC (VGG), Global=true, Regional=false2018.1194.7
R-MACNet=Res, F-tuned=yes, Dim=20482017.1194.5
Iscen's diffusionFeature=R-MAC (VGG), Global=true, Regional=false2018.1194.1
Proposed diffusionFeature=R-MAC (VGG), Global=false, Regional=true2018.1193.9
k-NN searchFeature=R-MAC (ResNet), Global=true, Regional=false2018.1193.8
GeMNet=Res, F-tuned=yes, Dim=20482017.1192.7
GeM+alphaQENet=VGG, F-tuned=yes, Dim=5122017.1191.9
R-MAC+QENet=VGG, F-tuned=yes, Dim=5122017.1191.2
R-MAC+QENet=Res, F-tuned=no, Dim=20482017.1189.7
k-NN + AQEFeature=R-MAC (VGG), Global=true, Regional=false2018.1188.4
GeMNet=VGG, F-tuned=yes, Dim=5122017.1187.7
R-MACNet=VGG, F-tuned=yes, Dim=5122017.1187.1
R-match + AQEFeature=R-MAC (VGG), Global=false, Regional=true2018.1187
R-MAC+AML+QE2015.1186.5
R-MAC+R+QENet=VGG, F-tuned=no, Dim=5122017.1186.5
R-matchFeature=R-MAC (VGG), Global=false, Regional=true2018.1186.1
R-MACNet=Res, F-tuned=no, Dim=20482017.1185.2
R-MACDimensionality=512, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop X2016.0485
CroW+QENet=VGG, F-tuned=no, Dim=5122017.1184.8
BOW-CNN+R+QENet=VGG, F-tuned=no, Dim=n/a2017.1184.8
k-NN searchFeature=R-MAC (VGG), Global=true, Regional=false2018.1184.5
HesAffNet-RootSIFTRetrieval Pipeline=+SV+QE(8)2017.1183.5
HesAffNet-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1183.3
HesAffNet-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1183.3
R-MACD=5122015.1183
R-MACNet=VGG, F-tuned=no, Dim=5122017.1183
HQENet=n/a, F-tuned=n/a, Dim=n/a2017.1182.8
Mikulik et al. (2013)2015.1182.4
Fisher VectorNet=VGG, F-tuned=yes, Dim=5122017.1182.4
BoW-fVocab+R+QENet=n/a, F-tuned=n/a, Dim=n/a2017.1182.4
BoW-CNNNet=VGG, F-tuned=no, Dim=n/a2017.1182
HesAffNet-RootSIFTRetrieval Pipeline=+SV+QE(15)2017.1181.7
ImageNet labelsBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.0781.5
Chum et al. (2011)2015.1180.5
BoW+R+QENet=n/a, F-tuned=n/a, Dim=n/a2017.1180.5
Danfeng et al. (2011)2015.1180.3
HesAff-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1179.9
CroWNet=VGG, F-tuned=no, Dim=5122017.1179.7
NetVLADNet=VGG, F-tuned=yes, Dim=40962017.1179.7
HesAffNetLess-HardNet++Retrieval Pipeline=+SV+QE(8)2017.1179.2
HesAff-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1179.1
HesAffNetLess-HardNet++Retrieval Pipeline=+SV+QE(15)2017.1179.1
SPOCNet=VGG, F-tuned=no, Dim=5122017.1178.2
MACDimensionality=128, Backbone=VGG, Training Protocol=Fine-tuned, Query Cropping=Crop2016.0477.6
NetVLADNet=VGG, F-tuned=no, Dim=40962017.1177.4
HesAff-RootSIFTRetrieval Pipeline=+SV+QE(8)2017.1177.4
Tolias et al. (2015)2015.1177
HesAff-RootSIFTRetrieval Pipeline=+SV+QE(15)2017.1176.4
NetVLADNet=VGG, F-tuned=yes, Dim=5122017.1174.9
HesAffNet-TFeat-M*Retrieval Pipeline=+SV+QE(8)2017.1174.8
CroWDimensionality=128, Backbone=VGG, Training Protocol=Off-the-shelf, Query Cropping=Crop2016.0474.6
Shen et al. (2014)2015.1174.1
HesAffNet-HardNet++Retrieval Pipeline=+SV2017.1173.4
R-MACD=2562015.1172.9
MACNet=VGG, F-tuned=no, Dim=5122017.1172.3
HesAffNet-TFeat-M*Retrieval Pipeline=+SV+QE(15)2017.1172
DeepClusterBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.0772
HesAffNet-RootSIFTRetrieval Pipeline=+SV2017.1171.2
HesAff-HardNet++Retrieval Pipeline=+SV2017.1170.3
HesAff-TFeat-M*Retrieval Pipeline=+SV+QE(8)2017.1169.7
HesAffNetLess-HardNet++Retrieval Pipeline=+SV2017.1168.7
Razavian et al. (2014b)D=2562015.1167
HesAffNet-HardNet++Retrieval Pipeline=BoW2017.1165.7
HesAff-TFeat-M*Retrieval Pipeline=+SV+QE(15)2017.1165.3
HesAff-HardNet++Retrieval Pipeline=BoW2017.1165
HesAff-RootSIFTRetrieval Pipeline=+SV2017.1163.7
HesAffNet-RootSIFTRetrieval Pipeline=BoW2017.1163.5
HesAffNetLess-HardNet++Retrieval Pipeline=BoW2017.1162
HesAff-RootSIFTRetrieval Pipeline=BoW2017.1159.3
HesAffNet-TFeat-M*Retrieval Pipeline=+SV2017.1158.1
Wang et al.Backbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=false2018.0758
Doersch et al.Backbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=false2018.0753.1
HesAff-TFeat-M*Retrieval Pipeline=+SV2017.1151.8
HesAffNet-TFeat-M*Retrieval Pipeline=BoW2017.1150.6
HesAff-TFeat-M*Retrieval Pipeline=BoW2017.1143.8
RandomBackbone=VGG-16, Retrieval Method=R-MAC, Resolution=1024, Grid levels=3, Sobel filtering=true2018.0722