Research map / Vision-language and multimodal / Cross-modal learning
Look, Listen and Learn
Relja Arandjelovic, Andrew Zisserman
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 AVC (L3) self-supervised training to the audio network (ablation) | L3-Net audio subnetwork · ESC-50 | accuracy (%): 62.5 -> 79.3 | Improved |
| add AVC (L3) self-supervised training to the audio network (ablation) | L3-Net audio subnetwork · DCASE | accuracy (%): 85 -> 93 | Improved |
| add AVC (L3) self-supervised training to the vision network (ablation) | L3-Net vision subnetwork · ImageNet linear classification | top-1 accuracy (%): 12.9 -> 32.3 | Improved |
| add AVC training to the vision network (ablation) | L3-Net vision subnetwork · Kinetics-Sounds clustering | NMI (vision): 0.204 -> 0.409 | Improved |
| add AVC training to the audio network (ablation) | L3-Net audio subnetwork · Kinetics-Sounds clustering | NMI (audio): 0.219 -> 0.330 | Improved |
| train both vision and audio networks instead of using fixed supervised teacher vision nets (SoundNet) | audio representation · ESC-50 | accuracy (%): 74.2 -> 79.3 | Improved |
| train both vision and audio networks instead of using fixed supervised teacher vision nets (SoundNet) | audio representation · DCASE | accuracy (%): 88 -> 93 | Improved |
| replace direct combination of supervised class scores with training the fusion layers on AVC | supervised baselines · Kinetics-Sounds AVC | AVC test accuracy (%): 65 -> 74 | Improved |
| replace supervised pretrained features with self-supervised L3 features | AVC fusion network · Kinetics-Sounds AVC | AVC test accuracy (%): 74 -> 74 | No clear effect |
| use L3-Net vision features instead of the earlier best self-supervised AlexNet features | ImageNet linear classification · ImageNet linear classification | top-1 accuracy (%): 34.7 -> 32.3 | Hurt |
Closest papers by meaning
- SoundNet: Learning Sound Representations from Unlabeled Video (NeurIPS 2016 · similarity 0.91)
- Self-Supervised Learning of Audio-Visual Objects from Video (ECCV 2020 · similarity 0.90)
- Objects that Sound (ECCV 2018 · similarity 0.90)
- Self-Supervised Object Detection From Audio-Visual Correspondence (CVPR 2022 · similarity 0.89)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.88)
- Separating the "Chirp" from the "Chat": Self-supervised Visual Grounding of Sound and Language (CVPR 2024 · 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.
