Research map / Vision-language and multimodal / Human and embodied modeling
Image2song: Song Retrieval via Bridging Image Content and Lyric Words
Xuelong Li, Di Hu, Xiaoqiang Lu
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 tag attention to lyric modeling (ablation) | Our baseline · Shuttersong image2song (dataset-S, obj-attr tags) | R@10: 4.94 -> 5.63 | Improved |
| add tag attention to lyric modeling (ablation) | Our baseline · Shuttersong image2song (dataset-dagger, obj-attr tags) | R@5: 7.6 -> 9.4 | Improved |
| add tag attention to lyric modeling (ablation) | Our baseline · Shuttersong image2song (dataset-dagger, obj-attr tags) | Med r: 43.30 -> 41.50 | Improved |
| use final LSTM output with tag attention instead of attentive pooling | Attentive-Reader · Shuttersong image2song (dataset-dagger, obj-attr tags) | R@10: 15.0 -> 16.8 | Improved |
| use bi-directional LSTM lyric encoder instead of BoW | BoW · Shuttersong image2song (dataset-dagger, obj-attr tags) | R@1: 1.4 -> 2.2 | Improved |
| use both object and attribute tags instead of attribute tags only (ablation) | Our-attention · Shuttersong image2song (dataset-dagger, obj-attr tags) | R@1: 2.2 -> 2.6 | Improved |
| average pooling instead of max pooling for tag attention (ablation) | Our-attention · Image2song pooling ablation (obj-attr tags) | R@1: 12.08 -> 13.10 | Improved |
| use MSE loss instead of marginal ranking loss (ablation) | Our-attention · Image2song loss ablation (obj-attr tags) | R@1: 9.90 -> 13.10 | Improved |
| use MSE loss instead of cosine proximity loss (ablation) | Our-attention · Image2song loss ablation (obj-attr tags) | R@1: 11.29 -> 13.10 | Improved |
| add mood tag to lyric representation (ablation) | Our · Image2song (obj-attr tags, supplementary) | R@1: 10.95 -> 12.13 | Improved |
| train with image-content supervision of lyric modeling instead of lyric pooling | Schwarz et al. · Song2image (lyric query, dataset-dagger) | R@1: 0.10 -> 0.19 | Improved |
Closest papers by meaning
- S²MILE: Semantic-and-Structure-Aware Music-Driven Lyric Generation (AAAI 2025 · similarity 0.85)
- SongGLM: Lyric-to-Melody Generation with 2D Alignment Encoding and Multi-Task Pre-Training (AAAI 2025 · similarity 0.84)
- MCSSME: Multi-Task Contrastive Learning for Semi-supervised Singing Melody Extraction from Polyphonic Music (AAAI 2024 · similarity 0.83)
- CSL-L2M: Controllable Song-Level Lyric-to-Melody Generation Based on Conditional Transformer with Fine-Grained Lyric and Musical Controls (AAAI 2025 · similarity 0.83)
- HarmonySet: A Comprehensive Dataset for Understanding Video-Music Semantic Alignment and Temporal Synchronization (CVPR 2025 · similarity 0.82)
- SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation (ICML 2025 · similarity 0.81)
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.
