Image-to-Text Retrieval on EIPM 200,000 products (test)
33.2Recall@1VLM2Vec + AGCL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VLM2Vec + AGCLArchitecture=LLM-based Baselines2026.04 | 33.2 | 54.3 | 62.1 | |
| VLM2VecArchitecture=LLM-based Baselines2026.04 | 30.9 | 52.2 | 60.5 | |
| LLM2CLIP + AGCLArchitecture=LLM-based Baselines2026.04 | 29.6 | 52.7 | 61.8 | |
| LLM2CLIPArchitecture=LLM-based Baselines2026.04 | 27.5 | 50.9 | 60.3 | |
| BLIPArchitecture=BERT-based Baselines2026.04 | 23.7 | 44.5 | 53.6 | |
| ALBEFArchitecture=BERT-based Baselines2026.04 | 22.1 | 43.1 | 52.5 | |
| CLIPArchitecture=BERT-based Baselines2026.04 | 20.9 | 42.1 | 51.7 |