Research map / Video / Video-language understanding
V-LynX: Token Interface Alignment for Video+X LLMs
Jungin Park, Jiyoung Lee, Kwanghoon Sohn
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 attention response alignment (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 81.0 -> 87.1 | Improved |
| add distribution regularization (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 86.2 -> 87.1 | Improved |
| add interface alignment stage before instruction tuning (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 77.3 -> 87.1 | Improved |
| add attention response alignment (ablation) | V-LynX-0.5B · ScanQA | EM@1: 23.5 -> 26.4 | Improved |
| raise vision-encoder LoRA rank from 8 to 64 (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 86.1 -> 87.1 | Improved |
| use audio-related videos instead of all as reference set (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 87.7 vs 87.1 | No clear effect |
| use only 563 3D clips as reference set (ablation) | V-LynX-0.5B · ScanQA | CIDEr: 87.8 vs 87.1 | No clear effect |
| replace PAVE with V-LynX | LLaVA-OV-7B · AVSD | CIDEr: +10.1 CIDEr | Improved |
| replace PAVE with V-LynX | LLaVA-OV-7B · ScanQA/SQA3D | parameters: 195.0M vs. 475.0M | Improved |
| apply V-LynX to Qwen2.5-VL-3B | Qwen2.5-VL-3B · SQA3D | EM@1: 15.1 -> 59.7 | Improved |
| apply V-LynX to InternVL-2.5-4B | InternVL-2.5-4B · SQA3D | EM@1: 44.0 -> 61.1 | Improved |
| replace PAVE-0.5B with V-LynX-0.5B | LLaVA-OV-0.5B · VideoMME | accuracy: +6.8% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PAVE-0.5B — PAVE: Patching and Adapting Video Large Language Models (CVPR 2025) · better on 15 of 15 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Aligned Better, Listen Better for Audio-Visual Large Language Models (ICLR 2025 · similarity 0.87)
- Empowering LLMs with Pseudo-Untrimmed Videos for Audio-Visual Temporal Understanding (AAAI 2025 · similarity 0.87)
- TC-LLaVA: Rethinking the Transfer of LLava from Image to Video Understanding with Temporal Considerations (AAAI 2025 · similarity 0.86)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.86)
- LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token (ICLR 2025 · similarity 0.85)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.85)
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.
