Knowledge-Based Visual Question Answering (Direct Answer) on A-OKVQA (test)
57.5AccuracyOurs + MCAN (LLaMA 2)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours + MCAN (LLaMA 2)Backbone=LLaMA 2, Shot Selection Strategy=MCAN, Number of Captions=9, Query Ensembles=52023.10 | 57.5 | |
| ProphetShot Selection Strategy=MCAN2023.10 | 55.7 | |
| Ours (LLaMA 2)Backbone=LLaMA 22023.10 | 55.4 | |
| Ours + MCAN (LLaMA)Backbone=LLaMA, Shot Selection Strategy=MCAN2023.10 | 55 | |
| Ours (LLaMA)Backbone=LLaMA2023.10 | 53.8 | |
| Unified-IO2023.10 | 45.2 | |
| GPV-22023.10 | 40.7 | |
| KRISP2023.10 | 27.1 | |
| ClipCap2023.10 | 25.9 | |
| ViLBERT2023.10 | 25.9 | |
| LXMERT2023.10 | 25.9 |