Video Question Answering on TVQA (test)
86.1AccuracySupervised SOTA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Supervised SOTAEvaluation protocol=supervised2023.12 | 86.1 | — | |
| FrozenBiLMZero-shot=false, Venue (Year)=NeurIPS (2022)2025.10 | 82 | — | |
| VINDLUZero-shot=false, Venue (Year)=CVPR (2023)2025.10 | 79 | — | |
| HEROZero-shot=false, Venue (Year)=EMNLP (2020)2025.10 | 74.24 | — | |
| HEROInput Modality=video+subtitle2020.05 | 73.61 | — | |
| STAGEInput Modality=video+subtitle2020.05 | 70.23 | — | |
| PAMNVideo Feat.=cpt2019.04 | 66.77 | — | |
| TVQAVideo Feat.=cpt2019.04 | 65.46 | — | |
| POVQAZero-shot=true, Venue (Year)=This work (2026)2025.10 | 64.7 | — | |
| PAMNVideo Feat.=img2019.04 | 64.61 | — | |
| ProViQEvaluation protocol=zero-shot2023.12 | 64.6 | — | |
| TVQAVideo Feat.=img2019.04 | 63.57 | — | |
| ViLAZero-shot=false, Venue (Year)=ECCV (2024)2025.10 | 63.4 | — | |
| TVQAVideo Feat.=reg2019.04 | 63.19 | — | |
| SeViLAZero-shot=false, Venue (Year)=NeurIPS (2023)2025.10 | 61.6 | — | |
| FrozenBiLMEvaluation protocol=zero-shot2023.12 | 59.7 | — | |
| FrozenBiLM (w/ speech)Zero-shot=true, Venue (Year)=NeurIPS (2022)2025.10 | 59.7 | — | |
| GPT-4V (via IG-VLM)Zero-shot=true, Venue (Year)=arXiv (2024)2025.10 | 57.8 | — | |
| InternVideoZero-shot=false, Venue (Year)=arXiv (2022)2025.10 | 57.2 | — | |
| BLIP-2Zero-shot=false, Venue (Year)=ICML (2023)2025.10 | 54.5 | — | |
| IG-VLM (LLaVA-1.6 34B)Zero-shot=true, Venue (Year)=IEEE Access (2024)2025.10 | 51.1 | — | |
| Goldfish–7B (vision+subs)Zero-shot=true, Venue (Year)=ECCV (2024)2025.10 | 46.94 | — | |
| HawkEyeFine-tuned=false2024.03 | 41.1 | — | |
| Q-ViDZero-shot=true, Venue (Year)=Findings ACL (2024)2025.10 | 41 | — | |
| VideoChat2Fine-tuned=false2024.03 | 40.6 | — | |
| VideoChat2 (reported by Q-ViD)Zero-shot=true, Venue (Year)=CVPR (2024)2025.10 | 40.6 | — | |
| VideoChat2 (our impl.)Fine-tuned=false, Implementation=Reproduced2024.03 | 40.15 | — | |
| SeViLA (reported by Q-ViD)Zero-shot=true, Venue (Year)=NeurIPS (2023)2025.10 | 38.2 | — | |
| Goldfish–7B (vision only)Zero-shot=true, Venue (Year)=ECCV (2024)2025.10 | 36.45 | — | |
| InternVideo (reported by Q-ViD)Zero-shot=true, Venue (Year)=arXiv (2022)2025.10 | 35.9 | — | |
| TimeChatFine-tuned=false2024.03 | 31.33 | — | |
| Longest Answer2019.04 | 30.41 | — | |
| FrozenBiLM (vision-only)Zero-shot=true, Venue (Year)=NeurIPS (2022)2025.10 | 29.7 | — | |
| CLIP-VIT-L/14Evaluation protocol=zero-shot2023.12 | 26.1 | — | |
| RandomEvaluation protocol=zero-shot2023.12 | 20 | — | |
| BERT QASamples for Multimodal Pretext=02021.01 | — | 72.7 | |
| HEROSamples for Multimodal Pretext=7.6M2021.01 | — | 73.6 | |
| HEROSamples for Multimodal Pretext=02021.01 | — | 70.3 | |
| MSANSamples for Multimodal Pretext=02021.01 | — | 71.1 | |
| STAGESamples for Multimodal Pretext=02021.01 | — | 70.2 | |
| TVQASamples for Multimodal Pretext=02021.01 | — | 68.5 | |
| VX2TEXTSamples for Multimodal Pretext=02021.01 | — | 75 |