Research map / Vision-language and multimodal / Cross-modal learning
How Far are We From Generating Missing Modalities With Foundation Models?
Guanzhou Ke, Bo Wang, Guoqing Chao, Weiming Hu, Shengfeng He
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 miner agent (ablation) | AFM^2 · VGGSound (70% missing) | FID (missing image): +71.12 | Improved |
| add verifier agent (ablation) | AFM^2 · VGGSound (70% missing) | FID (missing image): +13.17 | Improved |
| add miner agent (ablation) | AFM^2 · VGGSound (70% missing) | F1 (missing text): -2.57 | Improved |
| add verifier agent (ablation) | AFM^2 · VGGSound (70% missing) | PESQ (missing audio): -0.27 | Improved |
| add mining of object location (ablation) | Paradigm 3 (4o) · VGGSound (70% missing) | FID (missing image): 118.91 -> 88.62 | Improved |
| add mining of object location (ablation) | Paradigm 3 (4o) · VGGSound (70% missing) | MER (missing text): 0.69 -> 0.53 | Improved |
| mine objects only vs default prompt (ablation) | Paradigm 3 (4o) · VGGSound (70% missing) | FID (missing image): 118.91 -> 97.19 | Improved |
| add color on top of location mining (ablation) | Paradigm 3 (4o) · VGGSound (70% missing) | FID (missing image): 88.62 -> 98.52 | Hurt |
| increase generation candidates from N=5 to N=25 (ablation) | Paradigm 2 (IB) · VGGSound (30% missing) | FID (missing image): 197.17 -> 102.06 | Improved |
| increase generation candidates from N=5 to N=25 (ablation) | Paradigm 3 (4o) · VGGSound (30% missing) | FID (missing image): 129.79 -> 63.27 | Improved |
| increase generation candidates from N=5 to N=25 (ablation) | Paradigm 3 (4o) · VGGSound (30% missing) | MER (missing text): 0.61 -> 0.47 | Improved |
| increase candidates from N=5 to N=25 (inference time cost) (ablation) | Paradigm 2 (IB) · VGGSound (30% missing) | time (s): 17.47 -> 87.84 | Hurt |
Closest papers by meaning
- Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness (NeurIPS 2026 · similarity 0.90)
- Knowledge Bridger: Towards Training-Free Missing Modality Completion (CVPR 2025 · similarity 0.89)
- Gradient-Guided Modality Decoupling for Missing-Modality Robustness (AAAI 2024 · similarity 0.87)
- Mitigating Pervasive Modality Absence Through Multimodal Generalization and Refinement (AAAI 2025 · similarity 0.87)
- Probabilistic Conformal Distillation for Enhancing Missing Modality Robustness (NeurIPS 2024 · similarity 0.86)
- Flex-MoE: Modeling Arbitrary Modality Combination via the Flexible Mixture-of-Experts (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.
