Visual Question Answering on OK-VQA standard (test)
66.1VQA AccuracyPaLM-E
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PaLM-EBase Models=PaLM-E (526B)2025.04 | 66.1 | — | |
| UniRVQA (InstructBLIP-T5XL)Base Models=InstructBLIP-T5XL, K=5, Know. Source=GoogleSearch2025.04 | 61.57 | 66.79 | |
| ProphetBase Models=GPT-3 (175B)2025.04 | 61.11 | — | |
| UniRVQA (BLIP2-T5XL)Base Models=BLIP2-T5XL, K=5, Know. Source=GoogleSearch2025.04 | 60.9 | 64.21 | |
| RA-VQA-v2 (FLMR)Base Models=BLIP2-T5XL(∼3B), K=5, Know. Source=GoogleSearch2025.04 | 60.75 | 62.01 | |
| REVIVEBase Models=GPT-3 (175B), K=40, Know. Source=Wikipedia2025.04 | 58 | — | |
| FlamingoBase Models=Flamingo (80B)2025.04 | 57.8 | — | |
| PALIBase Models=PALI (15B)2025.04 | 56.5 | — | |
| InstructBLIP-T5XL (fine-tuned)Base Models=InstructBLIP-T5XL2025.04 | 55.5 | 60.47 | |
| BLIP2-T5XL (fine-tuned)Base Models=BLIP2-T5XL2025.04 | 52.73 | 55.11 | |
| RA-VQABase Models=T5-Large, K=5, Know. Source=GoogleSearch2025.04 | 51.22 | 55.77 | |
| TRiGBase Models=T5-Large, K=100, Know. Source=Wikipedia2025.04 | 50.5 | 54.73 | |
| PICaBase Models=GPT-3 (175B)2025.04 | 48 | — | |
| BLIP (fine-tuned)Base Models=BLIP2025.04 | 45.74 | 48.89 | |
| KAT-T5Base Models=T5-Large, K=40, Know. Source=Wikipedia2025.04 | 44.25 | — | |
| InstructBLIP-T5XL w/o fine-tunedBase Models=InstructBLIP-T5XL2025.04 | 41.54 | 44.07 | |
| MAVExKnow. Source=Wikipeida2025.04 | 39.2 | — | |
| BLIP (zero-shot)Base Models=BLIP2025.04 | 34.46 | 36.99 | |
| BLIP2-T5XL w/o fine-tuneBase Models=BLIP2-T5XL2025.04 | 11.6 | 12.49 |