Research map / Video / Action recognition
Object-Region Video Transformers
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Integrate object-region and dynamics information throughout video-transformer layers | MF · SomethingElse compositional | Top-1 accuracy: 60.2 -> 69.7 | Helped |
| Add ORViT object-region and dynamics streams | MViT-B · AVA-V2.2 | mAP: +1.1 | Helped |
| Add ORViT blocks | MF · Something-Something V2 | Top-1 accuracy: +1.4 | Helped |
| Add ORViT blocks | TimeSformer · Diving48 | Top-1 accuracy: +8.0% | Helped |
| Add ORViT blocks | MF-HR · Epic-Kitchens100 | Action accuracy: +1.2 | Helped |
| Add ODM to the single layer-2 ORViT block (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 66.7% to 68.8% | Helped |
| Apply ORViT early at layer 2 rather than at layer 12 (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 63.9 -> 66.7 | Helped |
| Repeat ORViT at layers 2, 7, and 11 instead of using one layer-2 block with ODM (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 68.8 -> 69.7 | Helped |
| Use Object-Region Attention without ODM (ablation) | MF · SomethingElse compositional | Top-1 accuracy: 60.2 -> 67.4 | Helped |
| Replace real object regions with zero-coordinate boxes (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 69.7 -> 60.4 | Hurt |
| Replace real object regions with fixed grid boxes (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 69.7 -> 60.9 | Hurt |
| Replace real object regions with random boxes (ablation) | ORViT MF · SomethingElse compositional | Top-1 accuracy: 69.7 -> 60.7 | Hurt |
Closest papers by meaning
- Video Action Transformer Network (CVPR 2019 · similarity 0.89)
- Generative Video Transformer: Can Objects be the Words? (ICML 2021 · similarity 0.89)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.88)
- Understanding Video Transformers via Universal Concept Discovery (CVPR 2024 · similarity 0.88)
- Vision Transformers Need Registers (ICLR 2024 · 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.