Research map / Vision-language and multimodal / Cross-modal learning
MIRROR: Aligning Semantic Relations from Language to Image via Gromov--Wasserstein
Hong-Han Wang, Yuntao Wang, Hu Ding
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 MIRROR SI-GW regularization | LLaVA-1.5-7B · GQA overall | accuracy: 61.16 -> 61.93 | Improved |
| add MIRROR | LLaVA-NeXT-7B · GQA overall | accuracy: 64.74 -> 65.56 | Improved |
| add MIRROR | LLaVA-1.5-7B · BLINK average | accuracy: 40.5 -> 42.9 | Improved |
| add MIRROR | LLaVA-NeXT-13B · BLINK average | accuracy: 35.8 -> 38.3 | Improved |
| add MIRROR | LLaVA-1.5-13B · GQA Relation | accuracy: 54.30 -> 54.43 | No clear effect |
| add MIRROR | LLaVA-NeXT-7B · VQAv2 | accuracy: 82.2 -> 82.1 | No clear effect |
| add MIRROR | LLaVA-1.5-7B · VQAv2 | accuracy: 79.3 -> 81.1 | Improved |
| add MIRROR | LLaVA models · POPE | accuracy: 85.7 -> 86.3 (1.5-7B) | Improved |
| add MIRROR | LLaVA models · RealWorldQA | accuracy: 57.8 -> 57.7 (NeXT-7B) | No clear effect |
| add MIRROR | LLaVA-1.5-7B · BLINK low-level tasks (reflection, depth) | accuracy: Depth 51.6 -> 54.8, Reflectance 26.9 unchanged | Mixed |
| add layer decoupling (ablation) | MIRROR · GQA / BLINK | training stability: diverged without it | Improved |
| add head selection (ablation) | MIRROR · BLINK | accuracy: 35.6 -> 42.9 | Improved |
Closest papers by meaning
- Remodeling Semantic Relationships in Vision-Language Fine-Tuning (AAAI 2026 · similarity 0.90)
- Topology-Aware Representation Alignment for Semi-Supervised Vision-Language Learning (NeurIPS 2026 · similarity 0.89)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · similarity 0.89)
- Large Vision-Language Models Get Lost in Attention (ICML 2026 · similarity 0.88)
- Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment (AAAI 2026 · 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.
