Research map / Continual learning / Domain and prompt-based
Audio-Visual Continual Test-Time Adaptation without Forgetting
Sarthak Kumar Maharana, Akshay Mehra, Bhavya Ramakrishna, Yunhui Guo, Guan-Ming Su
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 KL divergence over audio+visual statistics instead of norm-based distance (ablation) | AVReCAP retrieval · Kinetics50-C (visual corruptions) | mean accuracy (%): 49.47 -> 62.37 | Improved |
| use KL divergence over audio+visual statistics instead of norm-based distance (ablation) | AVReCAP retrieval · Kinetics50-2C (bimodal) | mean accuracy (%): 48.79 -> 50.29 | Improved |
| add visual statistics to KL retrieval (audio-only KL) (ablation) | AVReCAP retrieval · Kinetics50-C (visual corruptions) | mean accuracy (%): 60.52 -> 62.37 | Improved |
| add audio statistics to KL retrieval (visual-only KL) (ablation) | AVReCAP retrieval · Kinetics50-2C (bimodal) | mean accuracy (%): 49.28 -> 50.29 | Improved |
| increase distance threshold tau (ablation) | AVReCAP · Kinetics50 and VGGSound benchmarks | mean accuracy | Hurt |
| add selective parameter retrieval to READ-style fusion adaptation | AVReCAP vs READ · Kinetics50-C (visual corruptions) | mean accuracy (%): 49.46 -> 62.37 | Improved |
| add selective parameter retrieval to READ-style fusion adaptation | AVReCAP vs READ · VGGSound-C (audio corruptions) | mean accuracy (%): 18.13 -> 30.71 | Improved |
| add selective parameter retrieval to READ-style fusion adaptation | AVReCAP vs READ · VGGSound-2C (bimodal) forgetting on source test set | accuracy drop (%): 27.9% -> 2.9% | Improved |
| enlarge buffer budget from 300 to 400 (ablation) | AVReCAP · VGGSound-C (audio corruptions) | mean accuracy (%): 29.42 -> 30.61 | Improved |
| reduce buffer budget from infinity to 50 via merging (ablation) | AVReCAP · Kinetics50-C (visual corruptions) | mean accuracy (%): 62.37 -> 62.75 | No clear effect |
| reduce batch size to 8 (ablation) | AVReCAP · Kinetics50-C and Kinetics50-2C | mean accuracy | No clear effect |
| use element merging with tighter budget eta=200 (ablation) | AVReCAP · VGGSound-2C (bimodal) | mean accuracy | Improved |
Closest papers by meaning
- Continual Audio-Visual Sound Separation (NeurIPS 2024 · similarity 0.89)
- $\texttt{AVROBUSTBENCH}$: Benchmarking the Robustness of Audio-Visual Recognition Models at Test-Time (NeurIPS 2025 · similarity 0.87)
- Mixtures of Experts for Audio-Visual Learning (NeurIPS 2024 · similarity 0.87)
- Few-Shot Audio-Visual Class-Incremental Learning with Temporal Prompting and Regularization (AAAI 2025 · similarity 0.86)
- AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning (CVPR 2025 · similarity 0.86)
- Dynamic Cross-Modal Prompt Generation for Multimodal Continual Instruction Tuning (NeurIPS 2026 · 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.
