| Dataset | Language | # Hate | # Non-Hate | Resource |
|---|---|---|---|---|
| HateMM | English | 431 | 652 | BitChute |
| Dataset | Language | # Hateful | # Offensive | # Normal | Resource |
|---|---|---|---|---|---|
| MultiHateClip | English | 82 | 256 | 662 | YouTube |
| MultiHateClip | Chinese | 128 | 194 | 678 | Bilibili |
"Is there any hateful content in this video? Respond 'Yes' or 'No' and explain why."
| MLLMs | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|
| Closed-source | ||||
| Gemini-1.5-pro | 0.64380 | 0.42741 | 0.94642 | 0.58889 |
| Azure AI Video Indexer | ||||
| Open-source | ||||
| VideoChat2 | ||||
| VideoLLaMA2 (30Frames) | 0.62442 | 0.54811 | 0.30536 | 0.39221 |
| VideoLLaMA2-AV | 0.47166 | 0.40212 | 0.75622 | 0.52504 |
| LLaVA-Next-Video(Image-Text, 24Frames) | 0.55863 | 0.46252 | 0.67285 | 0.54820 |
| LLaVA-OneVision(Image-Text. 24Frames) | 0.65836 | 0.80198 | 0.18794 | 0.30451 |