Image Retrieval on ROxford
0.7656mAP (self)phi_*
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| phi_*Allocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7656 | 2.26 | — | 15.44 | — | — | — | — | |
| phi_*Allocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7656 | 2.26 | — | 11.46 | — | — | — | — | |
| AdvBCTAllocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7571 | 59.03 | 49.31 | 57.3 | 53.01 | — | — | — | |
| LCEAllocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7516 | 57.18 | 49.73 | 57.31 | 53.25 | — | — | — | |
| Hot-refreshAllocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.751 | 57.84 | 49.19 | 56.53 | 52.6 | — | — | — | |
| Hot-refreshAllocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7395 | 56.83 | 49.01 | 53.63 | 51.21 | — | — | — | |
| AdvBCTAllocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7393 | 60.33 | 49.65 | 56.64 | 52.83 | — | — | — | |
| LCEAllocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.7289 | 58.95 | 49.04 | 55.67 | 52.15 | — | — | — | |
| BCTAllocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.6893 | 57.52 | 47.71 | 54.59 | 50.91 | — | — | — | |
| BCTAllocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.6866 | 54.42 | 47.81 | 55.86 | 51.52 | — | — | — | |
| UniBCTAllocation type=Extended-backbone (data), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.6742 | 56.18 | 48.48 | 56.8 | 52.31 | — | — | — | |
| UniBCTAllocation type=Extended-backbone (class), Model_old=ResNet-18, Model_new=ResNet-502023.05 | 0.6637 | 58.51 | 47.29 | 55.51 | 51.06 | — | — | — | |
| AdvBCTAllocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6451 | 54.82 | 50.16 | 54.87 | 52.35 | — | — | — | |
| ODPP9% Data=phi1, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6446 | 0.4785 | — | — | — | — | — | — | |
| ODPP30% Data=phi2, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6446 | 0.5505 | — | — | — | — | — | — | |
| phi_*Allocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6385 | 1.2 | — | 7.19 | — | — | — | — | |
| phi_*Allocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6385 | 1.29 | — | 3.38 | — | — | — | — | |
| R50 (new)Allocation Type=Backbone Extension (R18→R50)2025.03 | 0.6359 | — | — | — | — | — | — | — | |
| NDPPAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.6298 | 0.4925 | 49.51 | 57.63 | 0.5326 | — | — | — | |
| AdvBCT9% Data=phi1, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6271 | 0.4595 | — | — | — | — | — | — | |
| AdvBCT30% Data=phi2, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6271 | 0.5442 | — | — | — | — | — | — | |
| ODPPAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.6246 | 0.494 | 49.5 | 58.41 | 0.5359 | — | — | — | |
| Independent Training100% Data=phi3, Backbone=ResNet-182025.03 | 0.6238 | — | — | — | — | — | — | — | |
| AdvBCTAllocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6213 | 52.31 | 49.55 | 58.09 | 53.45 | — | — | — | |
| NDPP9% Data=phi1, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6162 | 0.4688 | — | — | — | — | — | — | |
| NDPP30% Data=phi2, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.6162 | 0.545 | — | — | — | — | — | — | |
| AdvBCTAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.6112 | 0.4411 | 48.77 | 54.82 | 0.516 | — | — | — | |
| LCEAllocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6085 | 51.35 | 49.65 | 57.41 | 53.34 | — | — | — | |
| LCEAllocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6049 | 54.29 | 49.37 | 53.95 | 51.51 | — | — | — | |
| Hot-refreshAllocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.6031 | 51.68 | 48.06 | 47.26 | 47.57 | — | — | — | |
| RACTAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.601 | 0.4404 | 48.07 | 52.51 | 0.5016 | — | — | — | |
| BCT9% Data=phi1, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.5938 | 0.4668 | — | — | — | — | — | — | |
| BCT30% Data=phi2, 100% Data=phi3, Backbone=ResNet-182025.03 | 0.5938 | 0.5234 | — | — | — | — | — | — | |
| UniBCTAllocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.5907 | 57.82 | 48.09 | 54.78 | 51.17 | — | — | — | |
| BCTAllocation type=Extended-class, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.5894 | 53.43 | 48.33 | 52.79 | 50.41 | — | — | — | |
| UniBCTAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.5839 | 0.4596 | 47.91 | 56.07 | 0.5166 | — | — | — | |
| Hot-refreshAllocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.5815 | 50.05 | 47.78 | 52.5 | 50.03 | — | — | — | |
| NDPP9% Data=phi1, 30% Data=phi2, Backbone=ResNet-182025.03 | 0.5802 | 0.4642 | — | — | — | — | — | — | |
| NDPPAllocation Type=Data Extension (9%→30%)2025.03 | 0.5802 | 0.4642 | 50.87 | 59.44 | 0.548 | — | — | — | |
| Dual-TuningAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.5785 | 0.4666 | 47.72 | 55.71 | 0.5141 | — | — | — | |
| BCTAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.5758 | 0.4627 | 47.23 | 55.73 | 0.5111 | — | — | — | |
| UniBCTAllocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.5724 | 50.97 | 48.9 | 59.19 | 53.52 | — | — | — | |
| ODPPAllocation Type=Data Extension (9%→30%)2025.03 | 0.5645 | 0.4624 | 50.55 | 58.75 | 0.5432 | — | — | — | |
| BCTAllocation type=Extended-data, Model_old=ResNet-18, Model_new=ResNet-182023.05 | 0.5634 | 49.66 | 48.02 | 54.71 | 51.13 | — | — | — | |
| ODPP9% Data=phi1, 30% Data=phi2, Backbone=ResNet-182025.03 | 0.5607 | 0.4481 | — | — | — | — | — | — | |
| Independent Training30% Data=phi2, Backbone=ResNet-182025.03 | 0.5577 | — | — | — | — | — | — | — | |
| R18 (new)Allocation Type=Data Extension (9%→30%)2025.03 | 0.5577 | — | — | — | — | — | — | — | |
| SSPL-CEAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.5461 | 0.4587 | 46.84 | 55.33 | 0.5073 | — | — | — | |
| RACTAllocation Type=Data Extension (9%→30%)2025.03 | 0.5435 | 0.4155 | 49.33 | 49.27 | 0.4929 | — | — | — | |
| phi_oAllocation type=Extended-class, Model_old=ResNet-182023.05 | 0.5434 | — | — | — | — | — | — | — | |
| phi_oAllocation type=Extended-backbone (class), Model_old=ResNet-182023.05 | 0.5434 | — | — | — | — | — | — | — | |
| AdvBCT9% Data=phi1, 30% Data=phi2, Backbone=ResNet-182025.03 | 0.5408 | 0.4288 | — | — | — | — | — | — | |
| AdvBCTAllocation Type=Data Extension (9%→30%)2025.03 | 0.5408 | 0.4288 | 49.3 | 53.23 | 0.5119 | — | — | — | |
| Dual-TuningAllocation Type=Data Extension (9%→30%)2025.03 | 0.5397 | 0.4505 | 48.93 | 56.66 | 0.5251 | — | — | — | |
| BT^2Allocation Type=Backbone Extension (R18→R50)2025.03 | 0.5346 | 0.4545 | 46.46 | 53.98 | 0.4993 | — | — | — | |
| BT^2Allocation Type=Data Extension (9%→30%)2025.03 | 0.5293 | 0.4596 | 48.86 | 56 | 0.5218 | — | — | — | |
| BCT9% Data=phi1, 30% Data=phi2, Backbone=ResNet-182025.03 | 0.5116 | 0.4475 | — | — | — | — | — | — | |
| BCTAllocation Type=Data Extension (9%→30%)2025.03 | 0.5116 | 0.4475 | 47.75 | 55.34 | 0.5123 | — | — | — | |
| SSPL-CEAllocation Type=Data Extension (9%→30%)2025.03 | 0.509 | 0.4509 | 48.29 | 55.14 | 0.5149 | — | — | — | |
| UniBCTAllocation Type=Data Extension (9%→30%)2025.03 | 0.5004 | 0.4379 | 49.38 | 55.99 | 0.5247 | — | — | — | |
| phi_oAllocation type=Extended-data, Model_old=ResNet-182023.05 | 0.4915 | — | — | — | — | — | — | — | |
| phi_oAllocation type=Extended-backbone (data), Model_old=ResNet-182023.05 | 0.4915 | — | — | — | — | — | — | — | |
| SSPLAllocation Type=Data Extension (9%→30%)2025.03 | 0.4277 | 0.4207 | 44.52 | 50.36 | 0.472 | — | — | — | |
| Independent Training9% Data=phi1, Backbone=ResNet-182025.03 | 0.4182 | — | — | — | — | — | — | — | |
| R18 (old)Allocation Type=Data Extension (9%→30%)2025.03 | 0.4182 | — | — | — | — | — | — | — | |
| R18 (old)Allocation Type=Backbone Extension (R18→R50)2025.03 | 0.4182 | — | — | — | — | — | — | — | |
| SSPLAllocation Type=Backbone Extension (R18→R50)2025.03 | 0.4071 | 0.4159 | 41.75 | 49.91 | 0.4541 | — | — | — | |
| Image-ImageIntra-modal=true, Backbone=ViT-B/32, Latency (ms)=7 ± 12026.03 | — | — | — | — | — | 42.6 | — | — | |
| Image-ImageIntra-modal=true, Backbone=ViT-B/16, Latency (ms)=6 ± 12026.03 | — | — | — | — | — | 46.6 | — | — | |
| Image-ImageIntra-modal=true, Backbone=ViT-L/14, Latency (ms)=11 ± 12026.03 | — | — | — | — | — | 57.5 | — | — | |
| Image-ImageIntra-modal=true, Backbone=ViT-B/16-open, Latency (ms)=6 ± 12026.03 | — | — | — | — | — | 65.3 | — | — | |
| Image-ImageIntra-modal=true, Backbone=PE-Core-B-16, Latency (ms)=8 ± 12026.03 | — | — | — | — | — | 61.6 | — | — | |
| IsoCLIPIntra-modal=true, Backbone=ViT-B/32, Latency (ms)=7 ± 12026.03 | — | — | — | — | — | 47.2 | — | — | |
| IsoCLIPIntra-modal=true, Backbone=ViT-B/16, Latency (ms)=6 ± 12026.03 | — | — | — | — | — | 51.8 | — | — | |
| IsoCLIPIntra-modal=true, Backbone=ViT-L/14, Latency (ms)=11 ± 12026.03 | — | — | — | — | — | 66.4 | — | — | |
| IsoCLIPIntra-modal=true, Backbone=ViT-B/16-open, Latency (ms)=6 ± 12026.03 | — | — | — | — | — | 67.3 | — | — | |
| IsoCLIPIntra-modal=true, Backbone=PE-Core-B-16, Latency (ms)=8 ± 12026.03 | — | — | — | — | — | 67 | — | — | |
| OriginalBackbone=CLIP, ViT=B/32, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 42.4 | — | — | |
| OriginalBackbone=CLIP, ViT=L/14, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 57.1 | — | — | |
| OriginalBackbone=SigLIP, ViT=B/16, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 50.6 | — | — | |
| OriginalBackbone=SLIP, ViT=B/16, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 36.5 | — | — | |
| OriginalBackbone=SigLIP2, ViT=B/16, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 52.5 | — | — | |
| OriginalBackbone=DINOv3, ViT=L/16, Similarity=⟨I, I⟩2026.03 | — | — | — | — | — | 91.4 | — | — | |
| OTIBackbone=CLIP, ViT=B/32, Similarity=⟨T, I⟩2026.03 | — | — | — | — | — | 43 | — | — | |
| OTIBackbone=CLIP, ViT=L/14, Similarity=⟨T, I⟩2026.03 | — | — | — | — | — | 62.4 | — | — | |
| OTIBackbone=SigLIP, ViT=B/16, Similarity=⟨T, I⟩2026.03 | — | — | — | — | — | 55.2 | — | — | |
| OTIBackbone=SLIP, ViT=B/16, Similarity=⟨T, I⟩2026.03 | — | — | — | — | — | 36.4 | — | — | |
| OTI (I→T)Intra-modal=false, Backbone=ViT-B/32, Latency (ms)=1879 ± 352026.03 | — | — | — | — | — | 43 | — | — | |
| OTI (I→T)Intra-modal=false, Backbone=ViT-B/16, Latency (ms)=1856 ± 562026.03 | — | — | — | — | — | 49.9 | — | — | |
| OTI (I→T)Intra-modal=false, Backbone=ViT-L/14, Latency (ms)=1872 ± 912026.03 | — | — | — | — | — | 62.4 | — | — | |
| OTI (I→T)Intra-modal=false, Backbone=ViT-B/16-open, Latency (ms)=1836 ± 832026.03 | — | — | — | — | — | 69.5 | — | — | |
| OTI (I→T)Intra-modal=false, Backbone=PE-Core-B-16, Latency (ms)=3252 ± 352026.03 | — | — | — | — | — | 65.7 | — | — | |
| PCA←Backbone=CLIP, ViT=B/32, Similarity=⟨I←, I←⟩2026.03 | — | — | — | — | — | 51.4 | — | — | |
| PCA←Backbone=CLIP, ViT=L/14, Similarity=⟨I←, I←⟩2026.03 | — | — | — | — | — | 64.5 | — | — | |
| PCA←Backbone=SigLIP, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | — | — | — | — | — | 57.9 | — | — | |
| PCA←Backbone=SLIP, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | — | — | — | — | — | 43.6 | — | — | |
| PCA←Backbone=SigLIP2, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | — | — | — | — | — | 59.4 | — | — | |
| QUESTAttention=QUEST, Pre-training=iBOT, Backbone=ViT-B/162026.03 | — | — | — | — | — | — | 38.6 | 15.3 | |
| Standard AttentionAttention=Standard, Pre-training=iBOT, Backbone=ViT-B/162026.03 | — | — | — | — | — | — | 38.1 | 13.8 |