Research map / Vision-language and multimodal / Cross-modal learning
AdaMML: Adaptive Multi-Modal Learning for Efficient Video Recognition
Rameswar Panda, Chun-Fu (Richard) Chen, Quanfu Fan, Ximeng Sun, Kate Saenko, Aude Oliva, Rogerio Feris
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 LSTM to policy network (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 86.82 -> 88.17 | Improved |
| make per-modality decisions instead of joint decision (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 88.02 -> 88.17 | Improved |
| add efficiency loss (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 88.82 -> 88.17 | Mixed |
| use cost-based weights (lambda_audio 0.05) instead of equal cost weights (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 86.82 -> 88.17 | Improved |
| learned policy instead of random test-time selection (ablation) | AdaMML · ActivityNet (RGB+Audio) | mAP: 58.31 -> 73.91 | Improved |
| learned policy instead of random train+test selection (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 86.28 -> 88.17 | Improved |
| anneal temperature with factor 0.965 instead of 0 (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy: 87.40 -> 88.17 | Improved |
| vary penalty factor gamma over 0, 2, 5, 10 (ablation) | AdaMML · Kinetics-Sounds (RGB+Audio) | top-1 accuracy | No clear effect |
| use adaptive modality selection instead of weighted fusion of all modalities | weighted fusion · Kinetics-Sounds (RGB+Audio) | GFLOPs: 145.17 -> 76.45 | Improved |
| use adaptive modality selection instead of weighted fusion of all modalities | weighted fusion · ActivityNet (RGB+Audio) | mAP: 72.88 -> 73.91 | Improved |
| use RGBDiff instead of flow in policy network | AdaMML · Kinetics-Sounds (RGB+Flow+Audio) | top-1 accuracy: 88.54 -> 89.06 | Improved |
| use AdaMML instead of LiteEval | LiteEval · Kinetics-Sounds (vs LiteEval) | top-1 accuracy: 72.02 -> 88.17 | Improved |
Later papers that compared with this method
1 later accepted paper reports AdaMML as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- EgoAdapt: Adaptive Multisensory Distillation and Policy Learning for Efficient Egocentric Perception (ICCV 2025 · better on 3 of 3 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
- SimMLM: A Simple Framework for Multi-modal Learning with Missing Modality (ICCV 2025 · similarity 0.87)
- Towards Robust Multimodal Open-set Test-time Adaptation via Adaptive Entropy-aware Optimization (ICLR 2025 · similarity 0.85)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.85)
- Mixtures of Experts for Audio-Visual Learning (NeurIPS 2024 · similarity 0.85)
- Multimodal Representation Learning by Alternating Unimodal Adaptation (CVPR 2024 · similarity 0.85)
- Modality-Collaborative Test-Time Adaptation for Action Recognition (CVPR 2024 · 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.
