Research map / Video / Action recognition
Learning Representations from Audio-Visual Spatial Alignment
Pedro Morgado, Yi Li, Nuno Nvasconcelos
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 |
|---|---|---|---|
| Pretrain with audio-visual spatial alignment instead of AVC | R2+1D-18 audio-visual encoder · YT-360 AVSA-Bin | accuracy: 61.20 -> 69.97 | Improved |
| Pretrain with AVSA instead of AVC | R2+1D-18 · UCF101 | Video@1 accuracy: 69.68 -> 73.80 | Improved |
| Pretrain with AVSA instead of AVTS | R2+1D-18 · HMDB | Video@1 accuracy: 35.89 -> 37.66 | Improved |
| Add curriculum learning (AVC phase first) (ablation) | AVSA · YT-360 segmentation (context) | mIoU: 34.30 -> 35.83 | Improved |
| Use transformer attention instead of MLP in translation networks (ablation) | AVSA · YT-360 segmentation (context) | mIoU: 34.35 -> 35.83 | Improved |
| Use full ambisonics instead of mono audio (ablation) | AVSA · UCF101 | Clip@1 accuracy: 64.90 -> 68.52 | Improved |
| Use ambisonics instead of stereo audio (ablation) | AVSA · YT-360 AVSA-Bin | accuracy: 71.11 -> 69.97 | Hurt |
| Increase number of viewpoints from 4 to 8 (ablation) | AVSA · Semantic segmentation and UCF | mIoU / accuracy: 35.83 -> 34.84 mIoU; 68.52 -> 66.44 UCF | Hurt |
| Add spatial negatives to instance negatives (ablation) | AVSA · UCF101 | Clip@1 accuracy: 64.09 -> 68.52 | Improved |
| Add temporal negatives on top of spatial (ablation) | AVSA · Segmentation vs UCF | mIoU / accuracy: 35.83 -> 36.11 mIoU; 68.52 -> 65.77 UCF | Mixed |
| Train on hard negatives only (no easy-negative curriculum) (ablation) | AVSA · YT-360 segmentation | mIoU: 35.83 -> 20.97 | Hurt |
| Use transformer depth 2 instead of direct prediction (ablation) | AVSA · UCF101 | Clip@1 accuracy: 65.59 -> 68.52 | Improved |
Closest papers by meaning
- Aligning Audio-Visual Joint Representations with an Agentic Workflow (NeurIPS 2024 · similarity 0.89)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.89)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.88)
- Robust Audio-Visual Segmentation via Audio-Guided Visual Convergent Alignment (CVPR 2025 · similarity 0.88)
- Aligned Better, Listen Better for Audio-Visual Large Language Models (ICLR 2025 · similarity 0.88)
- Crab: A Unified Audio-Visual Scene Understanding Model with Explicit Cooperation (CVPR 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.
