Research map / Vision-language and multimodal / Cross-modal learning
Omnivore: A Single Model for Many Visual Modalities
Rohit Girdhar, Mannat Singh, Nikhila Ravi, Laurens van der Maaten, Armand Joulin, Ishan Misra
Check 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 |
|---|---|---|---|
| Jointly train one shared model on image, video, and RGBD datasets instead of modality-specific models | Swin-B · Kinetics-400 | top-1 accuracy: 80.6 -> 83.3 | Improved |
| Jointly train one shared model on image, video, and RGBD datasets instead of modality-specific models | Swin-B · SUN RGB-D | top-1 accuracy: 64.8 -> 65.4 | Improved |
| Jointly train one shared model on image, video, and RGBD datasets instead of modality-specific models | Swin-L · Kinetics-400 | top-1 accuracy: 83.1 -> 84.1 | Improved |
| Jointly train one shared model on image, video, and RGBD datasets instead of modality-specific models | Swin-L · SUN RGB-D | top-1 accuracy: 69.1 -> 67.1 | Hurt |
| Use an Omnivore pretrained representation for downstream finetuning | Swin-L · iNaturalist-2018 | classification accuracy: 81.7 -> 84.1 | Improved |
| Use an Omnivore pretrained representation for downstream finetuning | Swin-B · Something Something-v2 | top-1 accuracy: +0.5 | Improved |
| Initialize from ImageNet-21K and jointly finetune instead of training from scratch (ablation) | Omnivore · ImageNet-1K | top-1 accuracy: --0.7 | Hurt |
| Initialize from ImageNet-21K and jointly finetune instead of training from scratch (ablation) | Omnivore · Kinetics-400 | top-1 accuracy: --0.9 | Hurt |
| Initialize from ImageNet-21K and jointly finetune instead of training from scratch (ablation) | Omnivore · SUN RGB-D | top-1 accuracy: +0.9 | Improved |
| Increase SUN RGB-D replication from 50 to 100 (ablation) | Omnivore · SUN RGB-D | top-1 accuracy change: --2.1 | Hurt |
| Increase ImageNet-21K proportion from 0.1 to 1.0 (ablation) | Omnivore · Kinetics-400 | top-1 accuracy change: --4.5 | Hurt |
| Increase ImageNet-21K proportion from 0.1 to 1.0 (ablation) | Omnivore · SUN RGB-D | top-1 accuracy change: +2.0 | Improved |
Closest papers by meaning
- OmniSegmentor: A Flexible Multi-Modal Learning Framework for Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- 4M-21: An Any-to-Any Vision Model for Tens of Tasks and Modalities (NeurIPS 2024 · similarity 0.86)
- VITRIX-UniViTAR: Unified Vision Transformer with Native Resolution (NeurIPS 2025 · similarity 0.86)
- Same or Not? Enhancing Visual Perception in Vision-Language Models (CVPR 2026 · similarity 0.85)
- ViT-Lens: Towards Omni-modal Representations (CVPR 2024 · similarity 0.85)
- UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling (NeurIPS 2024 · 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.
