Research map / Detection and segmentation / CNN features
Dynamic Head: Unifying Object Detection Heads With Attentions
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 |
|---|---|---|---|
| Add scale-aware attention to the baseline (ablation) | ATSS · COCO val2017 | AP: +0.9 AP | Helped |
| Add spatial-aware attention to the baseline (ablation) | ATSS · COCO val2017 | AP: +2.4 AP | Helped |
| Add task-aware attention to the baseline (ablation) | ATSS · COCO val2017 | AP: +1.3 AP | Helped |
| Add scale- and spatial-aware attention (ablation) | ATSS · COCO val2017 | AP: +2.9 AP | Helped |
| Use the full scale-, spatial-, and task-aware DyHead (ablation) | ATSS · COCO val2017 | AP: +3.6 AP | Helped |
| Increase DyHead depth from 1 to 6 blocks (ablation) | ATSS · COCO val2017 | AP: 36.7 -> 42.6 | Helped |
| Increase DyHead depth from 6 to 8 blocks (ablation) | ATSS · COCO val2017 | AP: 42.6 -> 42.5 | No clear effect |
| Increase DyHead depth from 6 to 10 blocks (ablation) | ATSS · COCO val2017 | AP: 42.6 -> 42.3 | Hurt |
| Add DyHead to existing detectors | Faster R-CNN, RetinaNet, ATSS, FCOS, and RepPoints · COCO | AP: 1.2 ~ 3.2 AP | Helped |
| Use DyHead with Swin-L and extra data | DyHead · COCO test-dev | AP: 58.7 -> 60.6 | Helped |
Closest papers by meaning
- Fine-Grained Dynamic Head for Object Detection (NeurIPS 2020 · similarity 0.89)
- Rethinking Classification and Localization for Object Detection (CVPR 2020 · similarity 0.87)
- DynamicDet: A Unified Dynamic Architecture for Object Detection (CVPR 2023 · similarity 0.86)
- Dynamic Refinement Network for Oriented and Densely Packed Object Detection (CVPR 2020 · similarity 0.85)
- SdalsNet: Self-Distilled Attention Localization and Shift Network for Unsupervised Camouflaged Object Detection (AAAI 2025 · similarity 0.85)
- Training-Free Open-Ended Object Detection and Segmentation via Attention as Prompts (NeurIPS 2024 · similarity 0.84)
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.