Visual Dialogue on VisDial (Accuracy)
70.9VisDial AccuracyPALI-X-55B
Evaluation Results
| Method | Links | |
|---|---|---|
| PALI-X-55Bmode=per-task finetuning2023.07 | 70.9 | |
| MiniGPT-4Training Data Size=3.4k, Data Selection Strategy=Full Data2024.02 | 66.74 | |
| SELF-FILTERTraining Data Size=500, Data Selection Strategy=Ours2024.02 | 65.39 | |
| MiniGPT-4Training Data Size=500, Data Selection Strategy=w/o Diversity2024.02 | 60.75 | |
| MiniGPT-4Training Data Size=500, Data Selection Strategy=Random2024.02 | 55.35 | |
| Emu-Izero-shot=true, instruction_tuned=true, prompt_strategy=2-example text-only prompt2023.07 | 51.1 | |
| Flamingo-9Bzero-shot=true, prompt_strategy=2-example text-only prompt2023.07 | 48 | |
| Emuzero-shot=true, prompt_strategy=2-example text-only prompt2023.07 | 47.8 | |
| Emuzero-shot=true, prompt_strategy=Chain-of-Thought2023.07 | 47.4 | |
| Emu-Izero-shot=true, instruction_tuned=true2023.07 | 43 |