Image Retrieval on CUB-200 (test)
86.6R@1VPTSP
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VPTSPArch=ViT-S/16, Parameters=1.56M2024.02 | 86.6 | 52.7 | 91.7 | 94.8 | — | — | — | |
| HypArch=ViT-S/16, Parameters=30.2M2024.02 | 85.6 | — | 91.4 | 94.8 | — | — | — | |
| Full fine-tuning (PA)Arch=ViT-S/16, Parameters=30.2M2024.02 | 85.1 | 51.1 | 91.1 | 94 | — | — | — | |
| CLIP + EPBackbone=CLIP, Probing Method=Efficient Probing (EP), Pre-training Dataset=ImageNet-1K, Evaluation Protocol=Zero-shot2025.06 | 80.3 | — | 88 | 92.9 | 95.8 | 97.6 | 98.6 | |
| LOCORE-baseRe-ranking images count=1002025.03 | 78.3 | 64.8 | 91.9 | 98.2 | — | — | — | |
| SigLIP + EPBackbone=SigLIP, Probing Method=Efficient Probing (EP), Pre-training Dataset=ImageNet-1K, Evaluation Protocol=Zero-shot2025.06 | 76.7 | — | 85.7 | 92 | 95.7 | 97.8 | 98.8 | |
| CLIPBackbone=CLIP, Probing Method=Frozen, Evaluation Protocol=Zero-shot2025.06 | 75 | — | 84.3 | 91.3 | 95.3 | 97.5 | 98.7 | |
| LOCORE-smallRe-ranking images count=1002025.03 | 74.6 | 61 | 89.1 | 97.3 | — | — | — | |
| LOCORE-tinyRe-ranking images count=1002025.03 | 71.4 | 58.1 | 86.8 | 96.4 | — | — | — | |
| aQERe-ranking images count=1002025.03 | 70.9 | 56.9 | 78.8 | 84.7 | — | — | — | |
| PA (vanilla)Arch=R50, Parameters=26.5M2024.02 | 69.7 | 26.5 | 80 | 87 | — | — | — | |
| SSR RerankRe-ranking images count=1002025.03 | 69.4 | 54.2 | 79 | 86.1 | — | — | — | |
| NIRArch=R50, Parameters=26.5M2024.02 | 69.1 | — | 79.6 | — | — | — | — | |
| ProxyNCA++Arch=R50, Parameters=26.5M2024.02 | 69 | — | 79.8 | 87.3 | — | — | — | |
| Global descriptors2025.03 | 68.9 | 49.8 | 79.4 | 87.3 | — | — | — | |
| RRTRe-ranking images count=1002025.03 | 68.7 | 55.6 | 85 | 95.6 | — | — | — | |
| BEiTv2 + EPBackbone=BEiTv2, Probing Method=Efficient Probing (EP), Pre-training Dataset=ImageNet-1K, Evaluation Protocol=Zero-shot2025.06 | 68.2 | — | 78.7 | 86.7 | 92.3 | 95.9 | 97.8 | |
| DQERe-ranking images count=1002025.03 | 67 | 54.6 | 75.5 | 82 | — | — | — | |
| AQERe-ranking images count=1002025.03 | 66.9 | 58.4 | 76.8 | 82.7 | — | — | — | |
| XBM+PAArch=BN, Parameters=12M2024.02 | 65.8 | — | 75.9 | 84 | — | — | — | |
| iBOT + EPBackbone=iBOT, Probing Method=Efficient Probing (EP), Pre-training Dataset=ImageNet-1K, Evaluation Protocol=Zero-shot2025.06 | 62.1 | — | 73.7 | 82.4 | 89.7 | 94 | 96.8 | |
| SigLIPBackbone=SigLIP, Probing Method=Frozen, Evaluation Protocol=Zero-shot2025.06 | 60.8 | — | 72.1 | 81.6 | 88.8 | 93.5 | 96.4 | |
| MAE + EPBackbone=MAE, Probing Method=Efficient Probing (EP), Pre-training Dataset=ImageNet-1K, Evaluation Protocol=Zero-shot2025.06 | 57.1 | — | 69.5 | 79.5 | 87.5 | 93 | 96.2 | |
| BEiTv2Backbone=BEiTv2, Probing Method=Frozen, Evaluation Protocol=Zero-shot2025.06 | 57 | — | 68.2 | 78 | 85.8 | 90.9 | 94.2 | |
| iBOTBackbone=iBOT, Probing Method=Frozen, Evaluation Protocol=Zero-shot2025.06 | 51.8 | — | 64 | 74.7 | 83.2 | 90 | 94.1 | |
| MAEBackbone=MAE, Probing Method=Frozen, Evaluation Protocol=Zero-shot2025.06 | 16.3 | — | 23.7 | 33.3 | 45 | 58 | 70.8 | |
| CombEmbBit-length=24 bits2021.06 | — | 30.4 | — | — | — | — | — | |
| CombEmbBit-length=48 bits2021.06 | — | 33.7 | — | — | — | — | — | |
| CombEmbBit-length=72 bits2021.06 | — | 33.6 | — | — | — | — | — | |
| GPQBit-length=24 bits2021.06 | — | 16.7 | — | — | — | — | — | |
| GPQBit-length=48 bits2021.06 | — | 18.4 | — | — | — | — | — | |
| GPQBit-length=72 bits2021.06 | — | 19.2 | — | — | — | — | — |