Research map / Vision-language and multimodal / Cross-modal learning
Learning Visual Styles from Audio-Visual Associations
Tingle Li, Yichen Liu, Andrew Owens, Hang Zhao
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 |
|---|---|---|---|
| condition on audio instead of YAMNet class labels | audio-driven image stylization GAN · Into the Wild test | AMT score: 1.833 ± 0.042 -> 3.273 ± 0.046 | Improved |
| condition on audio instead of manual keyword labels | audio-driven image stylization GAN · Into the Wild test | FID: 38.066 -> 34.139 | Improved |
| condition on audio instead of manual keyword labels | audio-driven image stylization GAN · Into the Wild test | CLIP score: 0.236 -> 0.238 | No clear effect |
| use audio instead of a random style image (AdaIN) | audio-driven image stylization GAN · Into the Wild test | AVC: 0.812 -> 0.820 | Improved |
| use audio instead of keyword labels for sunny target scenes | audio-driven image stylization GAN · Into the Wild test, Sunny-to-Rainy | AVC: 0.827 -> 0.831 | Improved |
| add NCE contrastive loss instead of cycle-consistency (ablation) | audio-driven image stylization GAN · Into the Wild test (ablation) | FID: 35.244 -> 34.139 | No clear effect |
| add early-fusion discriminator instead of late fusion (ablation) | audio-driven image stylization GAN · Into the Wild test (ablation) | FID: 54.025 -> 34.139 | Improved |
| add identity loss (ablation) | audio-driven image stylization GAN · Into the Wild test (ablation) | FID: 41.019 -> 34.139 | Improved |
| add identity loss (ablation) | audio-driven image stylization GAN · Into the Wild test (ablation) | AVC: 0.810 -> 0.820 | Improved |
| initialize audio network with SeLaVi pre-training | audio-driven image stylization GAN · Into the Wild test (pre-training) | FID: 34.139 -> 32.882 | Improved |
| initialize audio network with Wav2CLIP pre-training instead of SeLaVi | audio-driven image stylization GAN · Into the Wild test (pre-training) | FID: 32.882 -> 30.334 | Improved |
Closest papers by meaning
- SoundBrush: Sound as a Brush for Visual Scene Editing (AAAI 2025 · similarity 0.89)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.87)
- Sounding that Object: Interactive Object-Aware Image to Audio Generation (ICML 2025 · similarity 0.87)
- Dynamic Derivation and Elimination: Audio Visual Segmentation with Enhanced Audio Semantics (CVPR 2025 · similarity 0.87)
- Unraveling Instance Associations: A Closer Look for Audio-Visual Segmentation (CVPR 2024 · similarity 0.86)
- Images that Sound: Composing Images and Sounds on a Single Canvas (NeurIPS 2024 · similarity 0.86)
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.
