Dense Audio Captioning on FTAR (test)
35.3METEORQwen2-Audio (TimePro-RL)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen2-Audio (TimePro-RL)Scale=7B, Evaluation Protocol=Post-Trained with TimePro-RL2026.04 | 35.3 | 39.8 | |
| Qwen2.5-Omni (TimePro-RL)Scale=7B, Evaluation Protocol=Post-Trained with TimePro-RL2026.04 | 33.9 | 40.7 | |
| Qwen2-AudioScale=7B, Evaluation Protocol=Finetuned2026.04 | 32.2 | 35 | |
| Qwen2.5-OmniScale=7B, Evaluation Protocol=Finetuned2026.04 | 31.3 | 35.2 | |
| Kimi-AudioScale=7B, Evaluation Protocol=Finetuned2026.04 | 31.2 | 32.7 | |
| Audio-Flaming2Scale=3B, Evaluation Protocol=Finetuned2026.04 | 25.7 | 12.7 | |
| TimeAudioScale=7B, Evaluation Protocol=Finetuned2026.04 | 20.4 | 37.4 | |
| Qwen2-AudioScale=7B, Evaluation Protocol=Zero-shot2026.04 | 11.2 | 3 | |
| Qwen2.5-OmniScale=7B, Evaluation Protocol=Zero-shot2026.04 | 10.5 | 10.4 |