Research map / Vision-language and multimodal / Cross-modal learning
ChronusOmni: Improving Time Awareness of Omni-Modal Large Language Models
Yijing Chen, Yihan Wu, Kaisi Guan, Wenhui Tan, Yuchen Ren, Yuyue Wang, Ruihua Song, Liyun Ru
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add temporal interleaved tokenization (text timestamps interleaved with video and audio) (ablation) | ChronusOmni · ChronusAV V2T | R@0.7: 6.80 -> 45.95 | Improved |
| add temporal interleaved tokenization (ablation) | ChronusOmni · ChronusAV A2T | R@0.7: 13.00 -> 79.85 | Improved |
| add SFT stage before GRPO (ablation) | ChronusOmni · ChronusAV T2A | CIDEr: 6.91 -> 34.30 | Improved |
| add GRPO stage after SFT (ablation) | ChronusOmni · ChronusAV V2T | R@0.7: 15.45 -> 45.95 | Improved |
| add GRPO stage (ablation) | ChronusOmni · ChronusAV V2A/A2V | CIDEr: V2A 1.24 -> 13.60 | Improved |
| add SFT stage (ablation) | ChronusOmni · ChronusAV V2T | R@0.7: 37.20 -> 45.95 | Improved |
| use 128 inference frames instead of 64 (ablation) | ChronusOmni · ChronusAV V2T | R@0.7: 45.95 -> 43.85 | Hurt |
| use 32 inference frames instead of 64 (ablation) | ChronusOmni · ChronusAV V2T | R@0.7: 45.95 -> 32.95 | Hurt |
| apply ChronusOmni to Qwen3-Omni comparison | ChronusOmni · ChronusAV V2T | R@0.5: 37.85 -> 63.15 | Improved |
| add temporal training to Ola | ChronusOmni · VisSpeech | WER: 12.3 -> 9.1 | Improved |
| add temporal training to Ola | ChronusOmni · LibriSpeech and Video-MME | WER / accuracy | No clear effect |
| evaluate vs LongVALE-LLM trained on LongVALE (zero-shot ChronusOmni) | ChronusOmni · LongVALE Omni-DVC | CIDEr: 7.9 -> 5.6 | Hurt |
Closest papers by meaning
- LongVALE: Vision-Audio-Language-Event Benchmark Towards Time-Aware Omni-Modal Perception of Long Videos (CVPR 2025 · similarity 0.90)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.89)
- LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning (NeurIPS 2026 · similarity 0.88)
- HourVideo: 1-Hour Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.88)
- TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
