Research map / Detection and segmentation / CNN features
Feature Aggregated Queries for Transformer-Based Video Object Detectors
Read the paperCheck 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 |
|---|---|---|---|
| Add FAQ query aggregation | Transformer-based detectors · ImageNet VID val | AP: 53.5 -> 55.8; 53.7 -> 56.1; 54.2 -> 58.0; 55.4 -> 60.1 | Helped |
| Add vanilla query aggregation (ablation) | Deformable-DETR · ImageNet VID val | mAP: 55.4 -> 56.7 | Helped |
| Replace vanilla aggregation with dynamic query aggregation (ablation) | Deformable-DETR · ImageNet VID val | mAP: 56.7 -> 58.5 | Helped |
| Add the extra loss to dynamic query aggregation (ablation) | Deformable-DETR with DQA · ImageNet VID val | mAP: 58.5 -> 60.1 | Helped |
| Use separate basic and dynamic query groups without linking them to inputs (ablation) | Deformable-DETR · ImageNet VID val | mAP: 55.4 -> 58.9 | Helped |
| Generate 2 or 4 dynamic queries per basic-query group instead of one (ablation) | FAQ mapping function M · ImageNet VID val | mAP: 60.1 -> 60.3; 60.1 -> 60.5 | Helped |
| Replace global pooling and projection with MLPs (ablation) | FAQ mapping function M · ImageNet VID val | mAP: 60.1 -> 59.3 | Hurt |
| Generate dynamic queries with multi-head cross-attention (ablation) | FAQ mapping function M · ImageNet VID val | mAP: 60.1 -> 58.2 | Hurt |
| Shuffle or randomly regroup the basic queries (ablation) | FAQ mapping function M · ImageNet VID val | mAP: 60.1 -> 60.0; 60.1 -> 60.1 | No clear effect |
| Use Transformer rather than cosine similarity for query aggregation (ablation) | FAQ query aggregation · ImageNet VID val | mAP: 58.7 -> 60.1 | Helped |
| Use simple learnable networks rather than cosine similarity for query aggregation (ablation) | FAQ query aggregation · ImageNet VID val | mAP: 58.7 -> 59.5 | Helped |
| Increase r from 1 to 4 (ablation) | FAQ dynamic query generation · ImageNet VID val | mAP: 58.7 -> 60.1 | Helped |
Closest papers by meaning
- Learning Dynamic Query Combinations for Transformer-based Object Detection and Segmentation (ICML 2023 · similarity 0.88)
- D2FANet: Enhancing Video Object Detection with Dual-Domain Feature Aggregation Network (CVPR 2026 · similarity 0.88)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.88)
- YOLOV: Making Still Image Object Detectors Great at Video Object Detection (AAAI 2023 · similarity 0.88)
- Context Enhanced Transformer for Single Image Object Detection in Video Data (AAAI 2024 · similarity 0.87)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · 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.