Image-to-Text Retrieval on Flickr30K (TR@1, TR@5, TR@10)
70.1TR@1Cosine
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CosineFeature Extractor=CLIP2025.06 | 70.1 | 89.1 | 93.3 | |
| EuclideanFeature Extractor=CLIP2025.06 | 48.2 | 63.7 | 68.7 | |
| EuclideanFeature Extractor=BLIP2025.06 | 47.7 | 79.7 | 88.8 | |
| MLP (Contr.)Feature Extractor=CLIP2025.06 | 42.6 | 72 | 81.4 | |
| MLP (Contr.)Feature Extractor=BLIP2025.06 | 33.3 | 64.3 | 75 | |
| Transformer (Contr.)Feature Extractor=CLIP2025.06 | 29.4 | 61.6 | 73.9 | |
| Transformer (Contr.)Feature Extractor=ConvNeXt/XLNet2025.06 | 3.4 | 13.5 | 20.5 | |
| CASTBudget=500, Selection Baseline Category=Ours2026.05 | 3.21 | 11.67 | 19.37 | |
| RandomBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 3.12 | 9.76 | 16.7 | |
| FASTBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 3.08 | 11.32 | 17.62 | |
| HerdingBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 2.7 | 9.22 | 15.42 | |
| GlisterBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 2.44 | 9.44 | 14.88 | |
| GradMatchBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 2.38 | 9.08 | 15.1 | |
| EL2NBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 2.18 | 8.84 | 14.3 | |
| CASTBudget=200, Selection Baseline Category=Ours2026.05 | 2.05 | 7.64 | 13.15 | |
| EntropyBudget=500, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1.92 | 6.38 | 10.44 | |
| FASTBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1.76 | 7.12 | 12.04 | |
| PreSelBudget=500, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 1.74 | 5.82 | 9.6 | |
| ViSABudget=500, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 1.72 | 5.58 | 9.16 | |
| RandomBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1.66 | 6.94 | 11.4 | |
| ProphetBudget=500, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 1.64 | 6.32 | 10.42 | |
| CASTBudget=100, Selection Baseline Category=Ours2026.05 | 1.5 | 6.15 | 10.45 | |
| Dyn-PruneBudget=500, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 1.5 | 6.38 | 10.68 | |
| GradMatchBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1.2 | 4.54 | 8.04 | |
| RandomBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1.18 | 4.2 | 7.48 | |
| FASTBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 1 | 4.04 | 7.02 | |
| Dyn-PruneBudget=200, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 1 | 3.84 | 6.78 | |
| GlisterBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.98 | 4.68 | 7.82 | |
| EL2NBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.98 | 4.16 | 7.44 | |
| ProphetBudget=200, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.92 | 3.72 | 6.18 | |
| GlisterBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.84 | 3.42 | 5.56 | |
| HerdingBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.84 | 4.06 | 7.14 | |
| EntropyBudget=200, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.84 | 3.06 | 5.3 | |
| GradMatchBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.8 | 2.76 | 4.58 | |
| ViSABudget=200, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.8 | 3.42 | 5.74 | |
| PreSelBudget=200, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.74 | 2.54 | 5.08 | |
| Dyn-PruneBudget=100, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.7 | 2.38 | 4.56 | |
| PreSelBudget=100, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.62 | 2.1 | 3.62 | |
| EL2NBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.58 | 2.8 | 5.36 | |
| ViSABudget=100, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.52 | 2.44 | 3.96 | |
| EntropyBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.5 | 1.82 | 3.48 | |
| HerdingBudget=100, Selection Baseline Category=Unimodal Selection Baselines2026.05 | 0.48 | 2.34 | 4.84 | |
| ProphetBudget=100, Selection Baseline Category=Multimodal Selection Baselines2026.05 | 0.44 | 2.44 | 4.12 | |
| Transformer (MSE)Feature Extractor=ResNet/BERT2025.06 | 0.1 | 0.01 | 0.2 |