Research map / Continual learning / Class-incremental
Hierarchical Augmentation and Distillation for Class Incremental Audio-Visual Video Recognition
Yukun Zuo, Hantao Yao, Liansheng Zhuang, Changsheng Xu
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 hierarchical augmentation module (HAM) (ablation) | baseline · HAD on AVE 3 phases | AIA: 74.8 -> 87.6 | Improved |
| add hierarchical logical distillation (HLD) (ablation) | baseline · HAD on AVE 3 phases | AIA: 74.8 -> 80.5 | Improved |
| add hierarchical correlative distillation (HCD) (ablation) | baseline · HAD on AVE 3 phases | AIA: 74.8 -> 78.2 | Improved |
| combine HLD and HCD into HDM (ablation) | HLD or HCD alone · HAD on AVE 3 phases | AIA: 80.5 -> 82.2 | Improved |
| add HDM to HAM (ablation) | HAM · HAD on AVE 3 phases | AIA: 87.6 -> 88.9 | Improved |
| add low-level modal augmentation (LMA) only (ablation) | baseline · HAD on AVE 3 phases | AIA: 74.8 -> 86.4 | Improved |
| add high-level video augmentation (HVA) only (ablation) | baseline · HAD on AVE 3 phases | AIA: 74.8 -> 87.0 | Improved |
| combine LMA and HVA (ablation) | LMA or HVA alone · HAD on AVE 3 phases | AIA: 87.0 -> 87.6 | Improved |
| add video-level and distribution-level logical distillation together (ablation) | SLD alone · HAD on AVE 3 phases | AIA: 80.3 -> 80.5 | Improved |
| add snippet-level and video-level correlative distillation together (ablation) | VCD alone · HAD on AVE 3 phases | AIA: 77.9 -> 78.2 | Improved |
| add audio to visual modality (HAD vs HAD-V) (ablation) | HAD · AVK-400 fully supervised | accuracy: 76.1 -> 84.8 | Improved |
| increase Gaussian augmentation lambda from 0 to 0.05 (ablation) | HAD · HAD on AVE 3 phases | AIA: 88.2 -> 88.9 | 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.
- il2m — IL2M: Class Incremental Learning With Dual Memory (ICCV 2019) · better on 9 of 9 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
- Few-Shot Incremental Learning via Foreground Aggregation and Knowledge Transfer for Audio-Visual Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Few-Shot Audio-Visual Class-Incremental Learning with Temporal Prompting and Regularization (AAAI 2025 · similarity 0.86)
- UWAV: Uncertainty-weighted Weakly-supervised Audio-Visual Video Parsing (CVPR 2025 · similarity 0.85)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.85)
- Unraveling Instance Associations: A Closer Look for Audio-Visual Segmentation (CVPR 2024 · similarity 0.84)
- $\texttt{AVROBUSTBENCH}$: Benchmarking the Robustness of Audio-Visual Recognition Models at Test-Time (NeurIPS 2025 · similarity 0.84)
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.
