Research map / Label-efficient and robust learning / Recognition and long-tail
Active Contrastive Learning of Audio-Visual Video Representations
Shuang Ma, Zhaoyang Zeng, Daniel McDuff, Yale Song
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 |
|---|---|---|---|
| replace random negative sampling with active sampling | XMoCo · UCF101 | top-1 accuracy: 74.1 -> 77.2 (+3.1) | Improved |
| replace random negative sampling with active sampling | XMoCo · HMDB51 | top-1 accuracy: 38.7 -> 40.6 (+1.9) | Improved |
| replace random negative sampling with active sampling | XMoCo · ESC50 | top-1 accuracy: 73.0 -> 79.2 (+6.2) | Improved |
| add the additional FC layers for cross-modal gradient flow (ablation) | CM-ACC · UCF101 | top-1 accuracy: 75.8 -> 77.2 | Improved |
| add the additional FC layers for cross-modal gradient flow (ablation) | CM-ACC · HMDB51 | top-1 accuracy: 39.1 -> 40.6 | Improved |
| add the additional FC layers for cross-modal gradient flow (ablation) | CM-ACC · ESC50 | top-1 accuracy: 77.3 -> 79.2 | Improved |
| use gradient embeddings rather than feature embeddings for active sampling (ablation) | CM-ACC · UCF101 | top-1 accuracy: 74.5 -> 77.2 (+2.7) | Improved |
| use gradient embeddings rather than feature embeddings for active sampling (ablation) | CM-ACC · HMDB51 | top-1 accuracy: 38.2 -> 40.6 (+2.4) | Improved |
| use gradient embeddings rather than feature embeddings for active sampling (ablation) | CM-ACC · ESC50 | top-1 accuracy: 75.1 -> 79.2 (+4.1) | Improved |
| replace random sampling with OHEM | contrastive pretraining · UCF101, M=32 | top-1 accuracy: 61.9 -> 50.2 (-11.7) | Hurt |
| replace random sampling with OHEM | contrastive pretraining · HMDB51, M=32 | top-1 accuracy: 33.1 -> 26.8 (-6.3) | Hurt |
| replace random sampling with active sampling | contrastive pretraining · UCF101, M=32 | top-1 accuracy: 61.9 -> 78.0 (+16.1) | Improved |
Closest papers by meaning
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.89)
- Robust Audio-Visual Instance Discrimination (CVPR 2021 · similarity 0.89)
- Audio-Visual Instance Discrimination with Cross-Modal Agreement (CVPR 2021 · similarity 0.88)
- Audio-Visual Contrastive Learning with Temporal Self-Supervision (AAAI 2023 · similarity 0.88)
- EquiAV: Leveraging Equivariance for Audio-Visual Contrastive Learning (ICML 2024 · similarity 0.86)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · 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.
