Video Question Answering on EgoSchema Zero-shot
81.8AccuracyGPT-4
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4Input Type=gt-narration, Zero-shot=true, Evaluation Subset=500 question subset2023.11 | 81.8 | |
| TOPA-Llama2-13BNumber of Frames=10, Evaluation Protocol=Zero-shot2024.05 | 51 | |
| GPT-4Input Type=text, Zero-shot=true2023.11 | 48.26 | |
| IG-VLM-Llava13BNumber of Frames=6, Evaluation Protocol=Zero-shot2024.05 | 47 | |
| TOPA-Llama3-8BNumber of Frames=10, Evaluation Protocol=Zero-shot2024.05 | 44.2 | |
| TOPA-Llama2-7BNumber of Frames=10, Evaluation Protocol=Zero-shot2024.05 | 41.2 | |
| IG-VLM-Llava7BNumber of Frames=6, Evaluation Protocol=Zero-shot2024.05 | 35.8 | |
| InternVideoInput Type=frame, Zero-shot=true2023.11 | 32.1 | |
| InternVideoNumber of Frames=8, Evaluation Protocol=Zero-shot2024.05 | 32.1 | |
| FrozenBiLMNumber of Frames=10, Evaluation Protocol=Zero-shot2024.05 | 26.9 | |
| SeViLANumber of Frames=32, Evaluation Protocol=Zero-shot2024.05 | 22.7 |