Research map / Detection and segmentation / Open-set and few-shot detection
DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object Detection
Siheng Wang, Yanshu Li, Bohan Hu, Zhengdao Li, HaiboZhan, Linshan Li, Weiming Liu, Ruizhi Qian, Guangxin Wu, Hao Zhang, Jifeng Shen, Piotr Koniusz and 3 more
Read the paperCheck 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 |
|---|---|---|---|
| add hierarchical concept pool (DHCP) to baseline (ablation) | DeCo-DETR baseline · OV-COCO | AP50 novel: 30.4 -> 36.6 | Improved |
| add gradient isolation (PD-DuGi) (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 36.6 -> 37.5 | Improved |
| add cosine annealed alignment/detection loss weighting (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 37.5 -> 41.3 | Improved |
| add fine-grained prototypes (M2=4800) on top of coarse ones (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 30.8 -> 41.3 | Improved |
| double fine-grained prototypes from 4800 to 9600 (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 41.3 -> 41.5 | No clear effect |
| increase decoder queries from 300 to 2000 (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 36.5 -> 41.3 | Improved |
| use 13B+ LVLM instead of 7B for region descriptions (ablation) | DeCo-DETR (LLaVA-1.5) · OV-COCO | AP50 novel: 30.1 -> 38.2 | Improved |
| scale LVLM beyond 13B (to Qwen2.5-VL 32B) (ablation) | DeCo-DETR · OV-COCO | AP50 novel: 38.2 -> 38.9 | No clear effect |
| remove text encoder at inference via prototype pool (vs Grounding DINO-T) | DeCo-DETR · OV-COCO | latency / AP50 novel: 280ms -> 135ms; 42.1 -> 41.3 AP novel | Mixed |
| DeCo-DETR vs best prior on OV-COCO C-OVD | DeCo-DETR · OV-COCO (C-OVD) | AP50 novel: 46.2 (BIRDet) vs 44.9 | Hurt |
Later papers that compared with this method
1 later accepted paper reports DeCo-DETR as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- CoT-PL: Chain-of-Thought Pseudo-Labeling for Open-Vocabulary Object Detection (ECCV 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- CAKE — CAKE: Category Aware Knowledge Extraction for Open-Vocabulary Object Detection (AAAI 2025) · better on 3 of 3 shared results
- CORA — CORA: Adapting CLIP for Open-Vocabulary Detection With Region Prompting and Anchor Pre-Matching (CVPR 2023) · better on 1 of 1 shared result
- Deformable DETR — Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Open-Det: An Efficient Learning Framework for Open-Ended Detection (ICML 2025 · similarity 0.88)
- OVOD-Agent: A Markov-Bandit Framework for Proactive Visual Reasoning and Self-Evolving Detection (CVPR 2026 · similarity 0.88)
- DetCLIPv3: Towards Versatile Generative Open-vocabulary Object Detection (CVPR 2024 · similarity 0.88)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.88)
- Simple Image-Level Classification Improves Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.88)
- Learning Background Prompts to Discover Implicit Knowledge for Open Vocabulary Object Detection (CVPR 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.
