Research map / Vision-language and multimodal / Cross-modal learning
Modality-Balanced Collaborative Distillation for Multi-Modal Domain Generalization
Xiaohan Wang, Zhangtao Cheng, Ting Zhong, Leiting Chen, Fan Zhou
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 EMA to the multi-modal baseline (ablation) | ERM · EPIC-Kitchens flow+audio | top-1 accuracy (%): 56.47 -> 60.94 | Improved |
| add AMD, GCC and distillation loss on top of EMA (ablation) | MBCD · EPIC-Kitchens flow+audio | top-1 accuracy (%): 60.94 -> 63.08 | Improved |
| use adaptive modality dropout instead of none (ablation) | MBCD · HAC flow+audio | top-1 accuracy (%): 57.43 -> 59.48 | Improved |
| use adaptive instead of fixed (0.5) modality dropout (ablation) | MBCD · HAC flow+audio | top-1 accuracy (%): 58.88 -> 59.48 | Improved |
| use fixed 0.5 dropout instead of none (ablation) | MBCD · HAC flow+audio | top-1 accuracy (%): 57.43 -> 58.88 | Improved |
| replace CMRF with MBCD | CMRF · EPIC-Kitchens video+audio+flow | top-1 accuracy (%): 65.47 -> 67.58 | Improved |
| replace CMRF with MBCD | CMRF · HAC video+audio+flow | top-1 accuracy (%): 70.64 -> 71.62 | Improved |
| replace CMRF with MBCD | CMRF · HAC flow+audio | top-1 accuracy (%): 57.74 -> 59.48 | Improved |
| replace CMRF with MBCD (H,A->C split) | CMRF · HAC video+audio+flow, H,A->C | top-1 accuracy (%): 55.61 -> 55.48 | No clear effect |
| add the third modality (flow) to MBCD vs best two-modality | MBCD · EPIC-Kitchens | top-1 accuracy (%): 63.08 -> 67.58 | Improved |
| add the third modality to ERM | ERM · EPIC-Kitchens | top-1 accuracy (%): 56.47 -> 50.55 | Hurt |
| replace CMRF with MBCD in-domain | CMRF · EPIC-Kitchens in-domain | top-1 accuracy (%): 79.82 -> 81.60 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SimMMDG — SimMMDG: A Simple and Effective Framework for Multi-modal Domain Generalization (NeurIPS 2023) · better on 7 of 7 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
- Cross-modal Representation Flattening for Multi-modal Domain Generalization (NeurIPS 2024 · similarity 0.91)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.88)
- Beyond Modality Collapse: Representation Blending for Multimodal Dataset Distillation (NeurIPS 2025 · similarity 0.87)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.86)
- Are We Making Progress in Multimodal Domain Generalization? A Comprehensive Benchmark Study (NeurIPS 2026 · similarity 0.86)
- Seeking Consistent Flat Minima for Better Domain Generalization via Refining Loss Landscapes (CVPR 2025 · 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.
