Research map / Video / Video-language understanding
Exploring Fine-Grained Audiovisual Categorization with the SSW60 Dataset
Grant Van Horn, Rui Qian, Kimberly Wilber, Hartwig Adam, Oisin Mac Aodha, Serge Belongie
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 |
|---|---|---|---|
| fuse audio and visual modalities (best fusion vs best single modality) | ViT-B audiovisual · SSW60 videos | top-1 accuracy: 76.2% -> 80.6% | Improved |
| pretrain visual backbone on Images iNat and audio on unpaired audio instead of ImageNet | ViT-B audiovisual fusion · SSW60 videos (fusion) | top-1 accuracy: 59.8% -> 80.6% | Improved |
| use score fusion instead of mid fusion (MBT) (ablation) | ViT-B audiovisual · SSW60 (iNat + Unpair Audio pretrain) | top-1 accuracy: 73.5% -> 80.6% | Improved |
| use mid fusion instead of late fusion (ImageNet pretrain) (ablation) | ViT-B audiovisual · SSW60 (ImageNet/ImageNet pretrain) | top-1 accuracy: 59.8% -> 54.3% | Hurt |
| fine-tune iNat-trained backbone on Video Frames | ResNet50 · SSW60 Video Frames | top-1 accuracy: 60.47 -> 71.88 | Improved |
| train on Images iNat instead of Video Frames directly | ResNet50 · SSW60 Video Frames | top-1 accuracy: 54.92 -> 60.47 | Improved |
| fine-tune unpaired audio model on Video Audio | ResNet18 · SSW60 Video Audio | top-1 accuracy: 24.41 -> 15.33 | Hurt |
| pretrain on unpaired audio instead of training on Video Audio only | ResNet18 · SSW60 Video Audio | top-1 accuracy: 10.37 -> 24.41 | Improved |
| add time crop augmentation (ablation) | ViT-B audio · SSW60 unpaired audio | top-1 accuracy: 44.1 -> 60.6 | Improved |
| add frequency mask augmentation (ablation) | ViT-B audio · SSW60 unpaired audio | top-1 accuracy: 60.6 -> 66.8 | Improved |
| use higher-resolution 384 ViT-B pretraining | ViT-B audio · SSW60 unpaired audio | top-1 accuracy: 66.8% -> 65.9% | Hurt |
| add AudioSet pretraining to ViT-B 384 | ViT-B audio · SSW60 unpaired audio | top-1 accuracy: 65.9% -> 67.4% | Improved |
Closest papers by meaning
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.86)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.86)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · similarity 0.85)
- Fine-Grained Perception in Panoramic Scenes: A Novel Task, Dataset, and Method for Object Importance Ranking (AAAI 2025 · similarity 0.85)
- Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis (CVPR 2025 · similarity 0.85)
- $\texttt{AVROBUSTBENCH}$: Benchmarking the Robustness of Audio-Visual Recognition Models at Test-Time (NeurIPS 2025 · similarity 0.85)
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.
