Image Retrieval on Flickr30k (Recall and Bias Metrics)
89.84Recall@5Baseline
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BaselineEvaluation protocol=finetuned2024.11 | 89.84 | — | — | — | — | — | — | |
| CLIP2023.05 | 89.1 | 0.115 | 0.315 | 0.3586 | 67.2 | 93.6 | — | |
| DAC2024.11 | 87.84 | — | — | — | — | — | — | |
| TSVLC (LLM+RB)Method variant=LLM+RB2024.11 | 87.46 | — | — | — | — | — | — | |
| DCRZero-shot=true2026.03 | 87.3 | — | — | — | — | — | — | |
| Original CLIPZero-shot=true2026.03 | 87.2 | — | — | — | — | — | — | |
| TSVLC (RB)Method variant=RB2024.11 | 87.12 | — | — | — | — | — | — | |
| MI-clip2023.05 | 85.4 | 0.096 | 0.2746 | 0.2951 | 63.9 | 91.3 | — | |
| PBM2023.05 | 85.3 | 0.036 | 0.1527 | 0.164 | 41.2 | 92.6 | — | |
| CLIPEvaluation protocol=standard2024.11 | 83.38 | — | — | — | — | — | — | |
| GenHancerZero-shot=true2026.03 | 81.6 | — | — | — | — | — | — | |
| NegCLIP2024.11 | 81.12 | — | — | — | — | — | — | |
| CLIPEvaluation protocol=finetuned2024.11 | 81.1 | — | — | — | — | — | — | |
| TripletCLIPEvaluation protocol=finetuned2024.11 | 80.88 | — | — | — | — | — | — | |
| CLIP-PGSImage Enc.=ViT-B/16, Ratio=0.3, Zero-shot=true2025.03 | 73.5 | — | — | — | 47.1 | 82 | — | |
| CLIP-PGSImage Enc.=ViT-B/16, Ratio=0.5, Zero-shot=true2025.03 | 72.9 | — | — | — | 45.3 | 81.2 | — | |
| E-CLIPImage Enc.=ViT-B/16, Zero-shot=true2025.03 | 70.9 | — | — | — | 43.3 | 80.2 | — | |
| CLIPImage Enc.=ViT-B/16, Zero-shot=true2025.03 | 70.4 | — | — | — | 43.2 | 80.4 | — | |
| A-CLIPImage Enc.=ViT-B/16, Zero-shot=true2025.03 | 70.1 | — | — | — | 43.1 | 78.8 | — | |
| SCAN2023.05 | 69.9 | 0.1098 | 0.3341 | 0.396 | 41.4 | 79.1 | — | |
| CLIP-PGSImage Enc.=ViT-S/16, Ratio=0.3, Zero-shot=true2025.03 | 69.1 | — | — | — | 42.7 | 78 | — | |
| CLIP-PGSImage Enc.=ViT-S/16, Ratio=0.5, Zero-shot=true2025.03 | 68.2 | — | — | — | 41.4 | 77.3 | — | |
| FLIPImage Enc.=ViT-B/16, Zero-shot=true2025.03 | 67.9 | — | — | — | 41.5 | 77.5 | — | |
| FairSample2023.05 | 67.5 | 0.0744 | 0.2699 | 0.3537 | 35.8 | 77.7 | — | |
| PTM-STNumber of distilled pairs=10002026.03 | 42.8 | — | — | — | 17.5 | 56.3 | — | |
| PTM-STPairs=500, Ratio=1.7%2026.03 | 40.5 | — | — | — | 16 | 54 | — | |
| PTM-STPairs=200, Ratio=0.7%2026.03 | 34.7 | — | — | — | 12.5 | 48.5 | — | |
| LoRSPairs=500, Ratio=1.7%2026.03 | 32.9 | — | — | — | 12.7 | 44.9 | — | |
| LoRSPairs=200, Ratio=0.7%2026.03 | 29.8 | — | — | — | 10.8 | 40 | — | |
| LoRSNumber of distilled pairs=10002026.03 | 29 | — | — | — | 9.8 | 41.6 | — | |
| PTM-STPairs=100, Ratio=0.3%2026.03 | 28.4 | — | — | — | 9.6 | 41.5 | — | |
| EDGENumber of distilled pairs=10002026.03 | 28.2 | — | — | — | 9.9 | 40.5 | — | |
| LoRSPairs=100, Ratio=0.3%2026.03 | 24.4 | — | — | — | 7.8 | 35.5 | — | |
| EDGEPairs=500, Ratio=1.7%2026.03 | 21 | — | — | — | 6.7 | 30.5 | — | |
| MTT-VLPairs=500, Ratio=1.7%2026.03 | 20.2 | — | — | — | 6.6 | 30 | — | |
| MTT-VLPairs=200, Ratio=0.7%2026.03 | 16 | — | — | — | 4.6 | 25.5 | — | |
| MTT-VLPairs=100, Ratio=0.3%2026.03 | 15.7 | — | — | — | 4.7 | 24.6 | — | |
| RANDPairs=500, Ratio=1.7%2026.03 | 10.5 | — | — | — | 2.4 | 17.4 | — | |
| K-CENTPairs=500, Ratio=1.7%2026.03 | 10.4 | — | — | — | 3.5 | 17.3 | — | |
| HERDPairs=500, Ratio=1.7%2026.03 | 10 | — | — | — | 3 | 17 | — | |
| FORGETPairs=500, Ratio=1.7%2026.03 | 9 | — | — | — | 1.8 | 15.9 | — | |
| HERDPairs=200, Ratio=0.7%2026.03 | 5.5 | — | — | — | 1.5 | 9.3 | — | |
| K-CENTPairs=200, Ratio=0.7%2026.03 | 5.4 | — | — | — | 1.5 | 9.9 | — | |
| RANDPairs=200, Ratio=0.7%2026.03 | 4.8 | — | — | — | 1.1 | 9.2 | — | |
| RANDPairs=100, Ratio=0.3%2026.03 | 4 | — | — | — | 1 | 6.5 | — | |
| K-CENTPairs=100, Ratio=0.3%2026.03 | 3.1 | — | — | — | 0.7 | 6.1 | — | |
| FORGETPairs=200, Ratio=0.7%2026.03 | 3.1 | — | — | — | 1.2 | 8.4 | — | |
| HERDPairs=100, Ratio=0.3%2026.03 | 2.8 | — | — | — | 0.7 | 5.3 | — | |
| FORGETPairs=100, Ratio=0.3%2026.03 | 2.4 | — | — | — | 0.7 | 5.6 | — | |
| SAFT-LAdversarial Attack=AutoAttack, Perturbation Budget (epsilon)=1/2552026.02 | — | — | — | — | — | — | 63.86 | |
| TGA-ZSRAdversarial Attack=AutoAttack, Perturbation Budget (epsilon)=1/2552026.02 | — | — | — | — | — | — | 61.14 |