Research map / Label-efficient and robust learning / Test-time adaptation
Bridging Modalities via Progressive Re-alignment for Multimodal Test-Time Adaptation
Jiacheng Li, Songhe Feng
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 |
|---|---|---|---|
| use L_PMGFA instead of KL-divergence for feature alignment (ablation) | BriMPR · Kinetics50-C (multimodal shift) | average accuracy (%): 31.5 -> 40.5 | Improved |
| use L_PMGFA instead of squared moment matching (ablation) | BriMPR · Kinetics50-C (multimodal shift) | average accuracy (%): 34.5 -> 40.5 | Improved |
| use non-squared instead of squared moment matching (ablation) | moment matching alignment · Kinetics50-C (multimodal shift) | average accuracy (%): 34.5 -> 37.4 | Improved |
| add inter-modal instance-wise contrastive learning L_IICL (ablation) | BriMPR · Kinetics50-C (multimodal shift) | average accuracy (%): 40.5 -> 40.7 | Improved |
| add cross-modal masked embedding recombination L_CMER (ablation) | BriMPR · Kinetics50-C (multimodal shift) | average accuracy (%): 40.7 -> 40.9 | Improved |
| add L_CMER (ablation) | BriMPR · VGGSound-C (audio corruption) | average accuracy (%): 35.3 -> 36.5 | Improved |
| use correct CMER weights instead of swapped lambda^a and lambda^v (ablation) | BriMPR · VGGSound-C (audio corruption) | average accuracy (%): 32.1 -> 36.5 | Improved |
| use correct CMER weights instead of swapped lambda^a and lambda^v (ablation) | BriMPR · Kinetics50-C (audio corruption) | average accuracy (%): 70.0 -> 72.0 | Improved |
| apply BriMPR when dominant modality (video) is corrupted | CAV-MAE · Kinetics50-C (video corruption) | average accuracy (%): 60.5 -> 65.9 | Improved |
| apply BriMPR when dominant modality (audio) is corrupted | CAV-MAE · VGGSound-C (audio corruption) | average accuracy (%): 25.0 -> 36.5 | Improved |
| apply BriMPR on real-world shift | Transformer · MOSI to SIMS | accuracy (%): 46.0 -> 58.2 | Improved |
| apply unimodal TTA Tent to audio corruption | CAV-MAE · VGGSound-C (audio corruption) | average accuracy (%): 25.0 -> 4.6 | Hurt |
Closest papers by meaning
- Towards Robust Multimodal Open-set Test-time Adaptation via Adaptive Entropy-aware Optimization (ICLR 2025 · similarity 0.91)
- Smoothing the Shift: Towards Stable Test-Time Adaptation under Complex Multimodal Noises (ICLR 2025 · similarity 0.90)
- Test-Time Selective Adaptation for Uni-Modal Distribution Shift in Multi-Modal Data (ICML 2025 · similarity 0.89)
- Dual Prototype Evolving for Test-Time Generalization of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.88)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.87)
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.
