Research map / Vision-language and multimodal / Cross-modal learning
Meric: A Unified Framework for Multimodal Music Generation and Retrieval via Representation Space Anchoring
Xihua Wang, Yinbo Wang, Jingchao Zhang, Ruihua Song
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 |
|---|---|---|---|
| replace CLIP ViT-H-14 with Qwen3VL-Embedding as vision encoder (ablation) | Music Head · MelBench | FADmert: 2.54 -> 1.95 | Improved |
| replace CLIP ViT-H-14 with Qwen3VL-Embedding as vision encoder (ablation) | Music Head · ARIA | FADmert: 4.73 -> 3.98 | Improved |
| use diffusion Music Head instead of linear MSE projector (ablation) | Meric projector · MelBench | FADmert: 4.37 -> 1.96 | Improved |
| use diffusion Music Head instead of contrastive linear projector (ablation) | Meric projector · MelBench | FADmert: 12.94 -> 1.96 | Improved |
| use diffusion Music Head instead of contrastive linear projector (ablation) | Meric projector · MelBench | R@1 / R@10: 1.70 / 9.02 -> 1.52 / 9.30 | Hurt |
| use diffusion Music Head instead of CrossVAE projector (ablation) | Meric projector · ARIA | FADmert: 7.22 -> 4.32 | Improved |
| use diffusion Music Head instead of BridgeVAE projector (ablation) | Meric projector · MelBench | FADmert: 4.06 -> 1.96 | Improved |
| use more than 1 DDIM step in Music Head sampling (ablation) | Music Head · Human listening study (100 images) | MOS: s=1 significantly worse (p<0.001); s>=5 indistinguishable (p>0.2) | Improved |
| increase Music Head DDIM steps from 5 to 20 (ablation) | Music Head · Human listening study (100 images) | MOS: p>0.2 | No clear effect |
| average predictions over N=10 seeds (ensemble) (ablation) | Music Head retrieval · natural-audio datasets (AudioSet-Music, MuImage, MelBench, EmoMV) | R@10: +4–10% | Improved |
| average predictions over N=10 seeds (ensemble) (ablation) | Music Head retrieval · ARIA | R@10 | No clear effect |
| replace Meric with ImageBind for image-music retrieval | retrieval baseline · MelBench | R@10 (I->M): 18.00 vs 8.62 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- CLAP — CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts (ECCV 2024) · better on 1 of 2 shared results
- ImageBind — ImageBind: One Embedding Space To Bind Them All (CVPR 2023) · better on 1 of 2 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
- MusicFlow: Cascaded Flow Matching for Text Guided Music Generation (ICML 2024 · similarity 0.87)
- UniMuMo: Unified Text, Music, and Motion Generation (AAAI 2025 · similarity 0.87)
- LLark: A Multimodal Instruction-Following Language Model for Music (ICML 2024 · similarity 0.86)
- MGE-LDM: Joint Latent Diffusion for Simultaneous Music Generation and Source Extraction (NeurIPS 2025 · similarity 0.86)
- JEN-1 Composer: A Unified Framework for High-Fidelity Multi-Track Music Generation (AAAI 2025 · similarity 0.85)
- CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction (ICML 2026 · 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.
