Research map / Video / Video-language understanding
VRDFormer: End-to-End Video Visual Relation Detection With 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 |
|---|---|---|---|
| Use recurrent queries to carry object-pair context across frames (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 27.25 -> 30.67 | Helped |
| Add re-activation for temporarily inactive recurrent queries (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 30.67 -> 32.43 | Helped |
| Jointly train object-pair detection and relation classification (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 29.85 -> 32.43 | Helped |
| Use 20 static queries instead of 100 (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 32.43 -> 21.67 | Hurt |
| Use 50 static queries instead of 100 (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 32.43 -> 30.12 | Hurt |
| Use 200 static queries instead of 100 (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 32.43 -> 28.62 | Hurt |
| Use 300 static queries instead of 100 (ablation) | VRDFormer · ImageNet-VidVRD | mAP: 32.43 -> 24.68 | Hurt |
| Replace temporal transformer self-attention with mean pooling (ablation) | VRDFormer relation classification · ImageNet-VidVRD | mAP: 32.43 -> 31.17 | Hurt |
| Replace temporal transformer self-attention with an LSTM (ablation) | VRDFormer relation classification · ImageNet-VidVRD | mAP: 32.43 -> 31.85 | Hurt |
| Reduce temporal aggregation length from 32 to 1 (ablation) | VRDFormer relation classification · ImageNet-VidVRD | mAP: 32.43 -> 30.57 | Hurt |
| Reduce temporal aggregation length from 32 to 4 (ablation) | VRDFormer relation classification · ImageNet-VidVRD | mAP: 32.43 -> 30.82 | Hurt |
| Reduce temporal aggregation length from 32 to 8 (ablation) | VRDFormer relation classification · ImageNet-VidVRD | mAP: 32.43 -> 31.25 | Hurt |
Closest papers by meaning
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.90)
- Beyond Short-Term Snippet: Video Relation Detection With Spatio-Temporal Global Context (CVPR 2020 · similarity 0.90)
- RelTransformer: A Transformer-Based Long-Tail Visual Relationship Recognition (CVPR 2022 · similarity 0.89)
- Groupwise Query Specialization and Quality-Aware Multi-Assignment for Transformer-based Visual Relationship Detection (CVPR 2024 · similarity 0.87)
- Multi-Modal Prompting for Open-Vocabulary Video Visual Relationship Detection (AAAI 2024 · similarity 0.87)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.86)
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.