Long Video Understanding on MLVU multiple-choice task
73.4Overall AccuracyLLaVA-Video + TV-RAG
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-Video + TV-RAG#Params=72B, Frames=642025.12 | 73.4 | |
| LLaVA-Video#Params=72B, Frames=64, replication=results of authors' replication2025.12 | 73.1 | |
| LLaVA-Video + TV-RAG#Params=7B, Frames=642025.12 | 72.6 | |
| Oryx-1.5#Params=32B, Frames=1282025.12 | 72.3 | |
| LLaVA-Video#Params=7B, Frames=64, replication=results of authors' replication2025.12 | 70.8 | |
| Aria#Params=25.3B, Frames=2562025.12 | 70.6 | |
| Video-XL#Params=7B, Frames=2562025.12 | 64.9 | |
| GPT-4oFrames=0.5 fps2025.12 | 64.6 | |
| Video-CCAM#Params=14B, Frames=962025.12 | 63.1 | |
| Video-TwGresolution=High-Resolution (HR)2026.02 | 60.3 | |
| Video-R1-7Breproduced=true2026.02 | 59.8 | |
| VideoChat-R1-7Breproduced=true2026.02 | 57 | |
| LongVA-7B2026.02 | 56.3 | |
| Qwen2.5-VL-7Bresolution=High-Resolution (HR)2026.02 | 55.3 | |
| Video-TwGresolution=Low-Resolution (LR)2026.02 | 54.6 | |
| VideoAgent2026.02 | 52.5 | |
| VideoTree2026.02 | 51.6 | |
| Qwen2.5-VL-7Bresolution=Low-Resolution (LR)2026.02 | 47.5 | |
| Video-LLaVA-7B2026.02 | 47.3 | |
| ShareGPT4Video-8B2026.02 | 46.4 | |
| Long-VILA-R1-7Breproduced=true2026.02 | 36.6 |