Research map / Detection and segmentation / CNN features
Region-based and query-based detectors
Design detectors using region proposals, learnable queries, and transformer pretraining.
183 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 Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
30 papers in 2023–24 and 22 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 9.1% to 6.3%.
Open questions the data points to
- Papers report Faster R-CNN on MS COCO val (AP (%)) anywhere from 19.3 to 27.2: comparisons against it depend on whose number you take.
No longer reported by papers here: COCO test-dev (AP, last in 2023); PASCAL VOC 2007 test (mAP, last in 2020); PASCAL VOC 2012 test (mAP, last in 2019); COCO minival (AP, last in 2021); COCO val (AP, last in 2024).
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 · You Only Look Once: Unified, Real-Time Object Detection (CVPR 2016 · 51,511 citations)
- 2017 · Feature Pyramid Networks for Object Detection (CVPR 2017 · 29,987 citations)
- 2018 · Single-Shot Refinement Neural Network for Object Detection (CVPR 2018 · 1,699 citations)
- 2019 · NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (CVPR 2019 · 2,060 citations)
- 2020 · EfficientDet: Scalable and Efficient Object Detection (CVPR 2020 · 9,058 citations)
- 2021 · Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021 · 1,852 citations)
- 2022 · Anchor DETR: Query Design for Transformer-Based Detector (AAAI 2022 · 484 citations)
- 2023 · YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors (CVPR 2023 · 11,788 citations)
- 2024 · Salience DETR: Enhancing Detection Transformer with Hierarchical Salience Filtering Refinement (CVPR 2024 · 119 citations)
- 2025 · YOLOv12: Attention-Centric Real-Time Object Detectors (NeurIPS 2025 · 637 citations)
- 2026 · Strip R-CNN: Large Strip Convolution for Remote Sensing Object Detection (AAAI 2026 · 22 citations)
Earliest papers here
- You Only Look Once: Unified, Real-Time Object Detection (CVPR 2016 · 51,511 citations)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · 3,426 citations)
- Training Region-Based Object Detectors With Online Hard Example Mining (CVPR 2016 · 2,755 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)
19 papers, 2017–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 36.2 | Feature Pyramid Networks for Object Detection (CVPR 2017 · 29,987 citations) |
| 2018 | 48.3 | An Analysis of Scale Invariance in Object Detection SNIP (CVPR 2018 · 936 citations) |
| 2020 | 55.1 | EfficientDet: Scalable and Efficient Object Detection (CVPR 2020 · 9,058 citations) |
| 2022 | 57.4 | R(Det)2: Randomized Decision Routing for Object Detection (CVPR 2022 · 9 citations) |
COCO val2017 · AP (higher is better)
14 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 40.3 | Bottom-Up Object Detection by Grouping Extreme and Center Points (CVPR 2019 · 1,062 citations) |
| 2020 | 42.8 | Seeing without Looking: Contextual Rescoring of Object Detections for AP Maximization (CVPR 2020 · 27 citations) |
| 2023 | 72.7 | Towards Efficient Use of Multi-Scale Features in Transformer-Based Object Detectors (CVPR 2023 · 53 citations) |
PASCAL VOC 2007 test · mAP (higher is better)
10 papers, 2016–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2016 | 80.5 | R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · 3,426 citations) |
| 2018 | 83.8 | Single-Shot Refinement Neural Network for Object Detection (CVPR 2018 · 1,699 citations) |
| 2019 | 84.1 | ScratchDet: Training Single-Shot Object Detectors From Scratch (CVPR 2019 · 149 citations) |
PASCAL VOC 2012 test · mAP (higher is better)
9 papers, 2016–2019 · not reported since 2019
| Year | Best so far | Set by |
|---|---|---|
| 2016 | 77.6 | R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · 3,426 citations) |
| 2018 | 83.5 | Single-Shot Refinement Neural Network for Object Detection (CVPR 2018 · 1,699 citations) |
| 2019 | 83.6 | ScratchDet: Training Single-Shot Object Detectors From Scratch (CVPR 2019 · 149 citations) |
COCO minival · AP (higher is better)
8 papers, 2017–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 33.9 | Feature Pyramid Networks for Object Detection (CVPR 2017 · 29,987 citations) |
| 2019 | 41.3 | Grid R-CNN (CVPR 2019 · 549 citations) |
| 2020 | 43.9 | Revisiting the Sibling Head in Object Detector (CVPR 2020 · 487 citations) |
COCO val · AP (higher is better)
8 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 40.9 | FreeAnchor: Learning to Match Anchors for Visual Object Detection (NeurIPS 2019 · 293 citations) |
| 2021 | 41.1 | IQDet: Instance-Wise Quality Distribution Sampling for Object Detection (CVPR 2021 · 68 citations) |
| 2022 | 43.6 | DESTR: Object Detection With Split Transformer (CVPR 2022 · 71 citations) |
| 2023 | 50.7 | End-to-End Multi-Object Detection with a Regularized Mixture Model (ICML 2023 · 0 citations) |
| 2024 | 60 | Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · 19 citations) |
What has been tried
163 papers here report 1,884 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 5%, hurt 18%, 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 |
|---|---|---|---|---|
| add area attention | CUDA latency, FP32 · latency (ms) | 2.7/2.5 -> 2.0/2.0; 5.1/4.4 -> 3.5/3.1; 26.4/21.9 -> 18.2/14.3 | Helped | YOLOv12: Attention-Centric Real-Time Object Detectors (NeurIPS 2025 · 637 citations) |
| Add mosaic augmentation | COCO val2017, 12 epochs · AP | 49.6 -> 50.4 | Helped | DEIM: DETR with Improved Matching for Fast Convergence (CVPR 2025 · 273 citations) |
| Add hierarchical query filtering with scale-independent salience supervision | CSD test · AP | 50.2 -> 52.0 | Helped | Salience DETR: Enhancing Detection Transformer with Hierarchical Salience Filtering Refinement (CVPR 2024 · 119 citations) |
| Apply one-to-many supervision to internal queries after cross-attention rather than output queries | COCO val2017 · mAP | 47.0 -> 47.6 | Helped | MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · 79 citations) |
| Add AMSP strategy to VConv | URPC · mAP@[0.5:0.05:0.95] | 0.397 -> 0.398 | Helped | AMSP-UOD: When Vortex Convolution and Stochastic Perturbation Meet Underwater Object Detection (AAAI 2024 · 58 citations) |
| jointly optimize with gaze box loss | GOO-Synth · AP | 83.7 -> 84.1 | Helped | TransGOP: Transformer-Based Gaze Object Prediction (AAAI 2024 · 24 citations) |
| Add MI to DINO | COCO val2017, ResNet-50, 12 epochs · AP | 49.0 -> 50.2 | Helped | MI-DETR: An Object Detection Model with Multi-time Inquiries Mechanism (CVPR 2025 · 19 citations) |
| Use Hungarian rather than IoU matching | COCO · AP | 38.3 -> 38.4 | No clear effect | Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · 19 citations) |
| Add one global image query | COCO · AP | 44.1 -> 45.0 | Helped | Frozen-DETR: Enhancing DETR with Image Understanding from Frozen Foundation Models (NeurIPS 2024 · 11 citations) |
| Use IoU instead of DIoU + epistemic uncertainty as the positive-anchor criterion | MS COCO · AP | 47.6 -> 44.6 | Hurt | Adaptive Important Region Selection with Reinforced Hierarchical Search for Dense Object Detection (NeurIPS 2024 · 2 citations) |
| Use intersection-based regression and intersection grouping rather than full-object regression | MS-COCO · mAP | 35.8 -> 38.1 | Helped | Union-over-Intersections: Object Detection beyond Winner-Takes-All (ICLR 2025 · 0 citations) |
| Add CLIP dictionary initialization instead of random initialization | COCO 2017 val · mAP | 50.02% -> 51.75% | Helped | YOLO-RD: Introducing Relevant and Compact Explicit Knowledge to YOLO by Retriever-Dictionary (ICLR 2025 · 0 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 |
|---|---|---|---|---|
| Faster R-CNN | MS COCO val | AP (%) | 19.3 – 27.2 | 2 papers, e.g. NeurIPS 2023 and NeurIPS 2016 |
| Mask R-CNN | COCO test-dev | AP | 36.9 – 39.8 | 2 papers, e.g. AAAI 2022 and CVPR 2019 |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MS COCO 2017 · 4 of 10 papers
- CrowdHuman · 2 of 10 papers
- COCO train2017 · 2 of 10 papers
Baselines
- YOLOv9 · 3 of 10 papers
- Faster R-CNN · 3 of 10 papers
- Deformable DETR · 3 of 10 papers
- RT-DETRv2 · 2 of 10 papers
- D-FINE · 2 of 10 papers
- YOLOv11 · 2 of 10 papers
- YOLOv10 · 2 of 10 papers
- YOLOv8 · 2 of 10 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 52 papers; from OpenAlex and the papers' affiliations).
- Beihang University · 4 papers
- Chinese Academy of Sciences · 4 papers
- Nanyang Technological University · 3 papers
- SenseTime · 3 papers
- Black Sesame Technologies · 2 papers
- Wenzhou University · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- The University of Sydney · 2 papers
- Zhejiang University · 2 papers
- Xi’an Jiaotong University · 2 papers
Use it
Download BibTeX of 40 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
- Box regression and loss design (94 papers)
- Efficient DETR query design (85 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).