Research map / Vision-language and multimodal / Cross-modal learning
MJEPA: A Simple and Scalable Joint-Embedding Predictive Architecture for Audio-Visual Learning
Revant Teotia, Adrien Bardes, Michael Rabbat, Sumit Chopra, Matthew Muckley, Nicolas Ballas
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 |
|---|---|---|---|
| share one encoder across audio and video without cross-modal alignment (ablation) | Unimodal baselines · AudioSet-20K (audio input) | mAP: 30.87 -> 28.70 | Hurt |
| share one encoder across audio and video without cross-modal alignment (ablation) | Unimodal baselines · AudioSet-20K (video input) | mAP: 19.84 -> 17.90 | Hurt |
| add cross-modal prediction losses to the shared encoder (ablation) | Shared encoder · AudioSet-20K (audio input) | mAP: 28.70 -> 33.52 | Improved |
| add cross-modal prediction losses to the shared encoder (ablation) | Shared encoder · AudioSet-20K (video input) | mAP: 17.90 -> 19.58 | Improved |
| add cross-modal prediction losses to the shared encoder (ablation) | Shared encoder · AudioSet-20K (audio-video input) | mAP: 32.38 -> 36.34 | Improved |
| add joint audio-video encoding with its intra and cross-modal losses (ablation) | Shared encoder + cross-modal alignment · AudioSet-20K (audio input) | mAP: 33.52 -> 38.89 | Improved |
| add joint audio-video encoding with its intra and cross-modal losses (ablation) | Shared encoder + cross-modal alignment · AudioSet-20K (video input) | mAP: 19.58 -> 25.38 | Improved |
| add joint audio-video encoding with its intra and cross-modal losses (ablation) | Shared encoder + cross-modal alignment · AudioSet-20K (audio-video input) | mAP: 36.34 -> 42.90 | Improved |
| add video-only VideoMix2M data (ablation) | MJEPA ViT-L · AudioSet-20K (audio input) | mAP: 38.89 -> 40.00 | Improved |
| add video-only VideoMix2M data (ablation) | MJEPA ViT-L · AudioSet-20K (video input) | mAP: 25.38 -> 29.63 | Improved |
| scale the encoder from ViT-L to ViT-g (ablation) | MJEPA + data scaling · AudioSet-20K (audio-video input) | mAP: 45.31 -> 45.44 | Improved |
| add AudioSet audio-visual data to video-only training (ablation) | MJEPA ViT-L · Kinetics-400 | top-1 accuracy (%): 80.6 -> 84.7 | Improved |
Closest papers by meaning
- Pushing the Frontier of Audiovisual Perception with Large-Scale Multimodal Correspondence Learning (CVPR 2026 · similarity 0.90)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.89)
- EquiAV: Leveraging Equivariance for Audio-Visual Contrastive Learning (ICML 2024 · similarity 0.88)
- CrossMAE: Cross-Modality Masked Autoencoders for Region-Aware Audio-Visual Pre-Training (CVPR 2024 · similarity 0.88)
- ES3: Evolving Self-Supervised Learning of Robust Audio-Visual Speech Representations (CVPR 2024 · similarity 0.88)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.88)
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.
