Research map / Vision-language and multimodal / Cross-modal learning
MODIX: A Training-Free Multimodal Information-Driven Positional Index Scaling for Vision-Language Models
Ruoxiang Huang, Zhen Yuan
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 MODIX positional stride scaling | Qwen3-VL-2B · ScienceQA | accuracy: 72.18 -> 78.28 | Improved |
| add MODIX positional stride scaling | LFM2-VL-1.6B · ScienceQA | accuracy: 65.41 -> 73.83 | Improved |
| add MODIX positional stride scaling | LFM2-VL-3B · DocVQA | exact match: 71.75 -> 79.33 | Improved |
| add MODIX positional stride scaling | Qwen3-VL-8B · BLINK | accuracy: 62.80 -> 61.05 | Hurt |
| add MODIX positional stride scaling | InternVL3.5-8B · RealWorldQA | accuracy: 63.79 -> 63.01 | No clear effect |
| add MODIX positional stride scaling | InternVL3.5-8B · DocVQA | exact match: 85.92 -> 85.63 | No clear effect |
| add MODIX positional stride scaling | Qwen3-VL-8B · Video-MME medium | accuracy: 58.79 -> 61.45 | Improved |
| use MODIX instead of CircleRoPE | Qwen3-VL-8B · ScienceQA | accuracy: 88.87 -> 90.16 | Improved |
| use MODIX instead of V2PE | InternVL3.5-8B · ScienceQA | accuracy: 89.27 -> 91.13 | Improved |
| use MODIX instead of MHRoPE | Qwen3-VL-8B · ChartQA | exact match: 72.67 -> 72.80 | Improved |
| fusion weight alpha 0.5 instead of 0.0 (add intra-modal term) (ablation) | MODIX Qwen3-VL-2B · ChartQA | exact match: 66.00 -> 68.76 | Improved |
| fusion weight alpha 0.5 instead of 1.0 (add inter-modal term) (ablation) | MODIX Qwen3-VL-2B · ScienceQA | accuracy: 76.90 -> 78.28 | Improved |
Closest papers by meaning
- Revisiting Multimodal Positional Encoding in Vision–Language Models (ICLR 2026 · similarity 0.91)
- Seeing is Understanding: Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs (ICML 2026 · similarity 0.90)
- DeepAlign: Mitigating Modality Conflict through Modality-Specific Alignment (CVPR 2026 · similarity 0.89)
- Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs (NeurIPS 2025 · similarity 0.89)
- Towards Long-window Anchoring in Vision-Language Model Distillation (AAAI 2026 · similarity 0.88)
- Lifting the Veil on Visual Information Flow in MLLMs: Unlocking Pathways to Faster Inference (CVPR 2025 · 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.
