Research map / Detection and segmentation / CNN features
Efficient DETR query design
Speed up and ease query competition in DETR-style detectors.
85 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in CNN features, keyword cluster “detectors · detr · search”
State History Benchmarks Tried Baselines Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
23 papers in 2023–24 and 22 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 7.0% to 6.3%.
No longer reported by papers here: COCO test-dev (AP, last in 2020).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Learning Activity Progression in LSTMs for Activity Detection and Early Detection (CVPR 2016 · 410 citations)
- 2017 · Learning Non-Maximum Suppression (CVPR 2017 · 712 citations)
- 2018 · Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · 7,059 citations)
- 2019 · Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression (CVPR 2019 · 6,284 citations)
- 2020 · NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · 242 citations)
- 2021 · $\alpha$-IoU: A Family of Power Intersection over Union Losses for Bounding Box Regression (NeurIPS 2021 · 145 citations)
- 2022 · DN-DETR: Accelerate DETR Training by Introducing Query DeNoising (CVPR 2022 · 1,031 citations)
- 2023 · Dense Distinct Query for End-to-End Object Detection (CVPR 2023 · 311 citations)
- 2024 · DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · 4,429 citations)
- 2025 · UniSite: The First Cross-Structure Dataset and Learning Framework for End-to-End Ligand Binding Site Detection (NeurIPS 2025 · 2 citations)
- 2026 · RF-DETR: Neural Architecture Search for Real-Time Detection Transformers (ICLR 2026 · 4 citations)
Earliest papers here
- Learning Activity Progression in LSTMs for Activity Detection and Early Detection (CVPR 2016 · 410 citations)
- Joint Training of Cascaded CNN for Face Detection (CVPR 2016 · 234 citations)
- Learning Non-Maximum Suppression (CVPR 2017 · 712 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
COCO test-dev · AP (higher is better)
4 papers, 2018–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 42.8 | Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · 7,059 citations) |
| 2020 | 46.1 | NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · 242 citations) |
What has been tried
65 papers here report 765 measured changes (a component added, removed or swapped, compared with a baseline): helped 70%, no clear effect 8%, hurt 20%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace the multi-scale Transformer encoder with the efficient hybrid encoder | COCO val2017, 1× configuration · AP | 46.4 -> 47.9 | Helped | DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · 4,429 citations) |
| use NMS-free training with dual assignments | COCO val · latency | 7.07 -> 2.44 | Helped | YOLOv10: Real-Time End-to-End Object Detection (NeurIPS 2024 · 1,586 citations) |
| add the scale weight w to the IoU-based scale loss | IRSTD-1k · IoU (%) | 64.83 -> 65.82 | Helped | Infrared Small Target Detection with Scale and Location Sensitivity (CVPR 2024 · 253 citations) |
| Add attention query/key representations alongside final hidden states | Visual Genome · R@50 | 29.9 -> 30.2 | Helped | EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · 64 citations) |
| Use DCUS instead of entropy sampling for candidate-pool selection | Pascal VOC, 7.5% labeled images · mAP | 59.0±0.6 -> 60.8±0.5 | Helped | Plug and Play Active Learning for Object Detection (CVPR 2024 · 35 citations) |
| Add oriented loss L_o to progressively organize latent-space anchors | C-Avg. cross-dataset evaluation · AUC | 0.8087 -> 0.8553 | Helped | Can We Leave Deepfake Data Behind in Training Deepfake Detector? (NeurIPS 2024 · 25 citations) |
| add global loss to local loss without filtering | COCO W2A4 · mAP | 23.0 -> 23.5 | Helped | Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector (CVPR 2024 · 17 citations) |
| Add object-level class-wise feature alignment to image-level alignment | COCO → COCO-C · mAP | 16.7 -> 22.6 | Helped | What How and When Should Object Detectors Update in Continually Changing Test Domains? (CVPR 2024 · 14 citations) |
| Add association training to Deformable DETR | CrowdHuman · AP face | 72.2 -> 72.6 | Helped | PairDETR : Joint Detection and Association of Human Bodies and Faces (CVPR 2024 · 8 citations) |
| replace EnCodec architecture with DPRNN | 1 second watermark in 10 second audio · IoU | 0.802 -> 0.796 | No clear effect | Proactive Detection of Voice Cloning with Localized Watermarking (ICML 2024 · 7 citations) |
| Add MSelf-attention | COCO 2017 val, 12 epochs, R50 · AP | 47.0 -> 48.1; +1.1 AP | Helped | EASE-DETR: Easing the Competition among Object Queries (CVPR 2024 · 6 citations) |
| Add selective context | Ox5k · mAP | 80.2 -> 87.6 | Helped | On Train-Test Class Overlap and Detection for Image Retrieval (CVPR 2024 · 4 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| H-DETR | COCO val2017 R50 12ep | AP | 45.9 – 48.7 | 2 papers, e.g. NeurIPS 2023 and NeurIPS 2023 |
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 45 papers; from OpenAlex and the papers' affiliations).
- Peking University · 5 papers
- Tsinghua University · 4 papers
- Tencent · 3 papers
- Baidu · 3 papers
- The Chinese University of Hong Kong · 2 papers
- Stanford University · 2 papers
- Zhejiang University · 2 papers
- A*STAR · 2 papers
- Beihang University · 2 papers
- Korea University · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Region-based and query-based detectors (183 papers)
- Box regression and loss design (94 papers)
- Pedestrian and crowd detection (48 papers)
Elsewhere in the topic
- Context modeling for semantic segmentation (241 papers)
- Local features and context in CNNs (170 papers)
- Context for hard detection conditions (133 papers)
- Domain-specific segmentation and detection (118 papers)
- Convolution-transformer backbones (101 papers)
- Boundary-aware instance segmentation (83 papers)
- Plain and high-resolution ViTs (69 papers)
- Arbitrary-shape scene text detection (61 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).