Image-Text Retrieval on Flickr30K (R@1, R@5, nDCG@10)
82.4Recall@1SaMer (w/ FT)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SaMer (w/ FT)Architecture=Multi-vector, Backbone=ColPali, Token Compression (K)=64, Compression Strategy=Merging, Adapted=true2026.07 | 82.4 | 96.3 | 90.9 | |
| EVA-CLIPArchitecture=Single-vector2026.07 | 79.9 | 94.7 | 89.2 | |
| SigLIP2Architecture=Single-vector2026.07 | 79.8 | 94.2 | 88.8 | |
| DFN-CLIPArchitecture=Single-vector2026.07 | 79.6 | 94.9 | 89 | |
| SaMer (w/ FT)Architecture=Multi-vector, Backbone=ColQwen2, Token Compression (K)=64, Compression Strategy=Merging, Adapted=true2026.07 | 79.3 | 94.7 | 88.7 | |
| ColPaliArchitecture=Multi-vector, Backbone=ColPali, Compression=None2026.07 | 77 | 92.9 | 86.9 | |
| OpenCLIPArchitecture=Single-vector2026.07 | 76.5 | 93.5 | 87 | |
| GMEArchitecture=VLM-based2026.07 | 74.6 | 91.8 | 85.2 | |
| H-PoolArchitecture=Multi-vector, Backbone=ColPali, Token Compression (K)=64, Compression Strategy=Pooling2026.07 | 73.7 | 91.4 | 84.7 | |
| ColQwen2Architecture=Multi-vector, Backbone=ColQwen2, Compression=None2026.07 | 73.6 | 91.3 | 84.7 | |
| SaMer (w/o FT)Architecture=Multi-vector, Backbone=ColPali, Token Compression (K)=64, Compression Strategy=Merging, Adapted=false2026.07 | 73.5 | 91.1 | 84.5 | |
| H-PoolArchitecture=Multi-vector, Backbone=ColQwen2, Token Compression (K)=64, Compression Strategy=Pooling2026.07 | 71.8 | 90.8 | 83.6 | |
| SaMer (w/o FT)Architecture=Multi-vector, Backbone=ColQwen2, Token Compression (K)=64, Compression Strategy=Merging, Adapted=false2026.07 | 71.3 | 90.4 | 83.3 | |
| VisRAGArchitecture=VLM-based2026.07 | 69.6 | 90 | 82.2 | |
| SAPArchitecture=Multi-vector, Backbone=ColQwen2, Token Compression (K)=64, Compression Strategy=Structural anchor pruning2026.07 | 69.3 | 89.9 | 82.2 | |
| SAPArchitecture=Multi-vector, Backbone=ColPali, Token Compression (K)=64, Compression Strategy=Structural anchor pruning2026.07 | 68.3 | 88.6 | 80.8 | |
| HPCArchitecture=Multi-vector, Backbone=ColQwen2, Token Compression (K)=64, Compression Strategy=Attention-based pruning2026.07 | 67.1 | 88.7 | 80.4 | |
| MetaCLIPArchitecture=Single-vector2026.07 | 62.8 | 85.8 | 77.2 | |
| CLIPArchitecture=Single-vector2026.07 | 58.2 | 82.9 | 73.7 | |
| HPCArchitecture=Multi-vector, Backbone=ColPali, Token Compression (K)=64, Compression Strategy=Attention-based pruning2026.07 | 54 | 77.8 | 69 | |
| VLM2Vec-v2Architecture=VLM-based2026.07 | 41.7 | 68.8 | 59 |