Research map / Detection and segmentation / CNN features
Box regression and loss design
Improve localization with refined losses and anchor-free or rotated box formulations.
94 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
Quiet None of the tracked benchmarks has a report from 2025–26.
21 papers in 2023–24 and 15 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.4% to 4.3%.
No longer reported by papers here: COCO minival (AP, last in 2023); COCO test-dev (AP, last in 2022); MS COCO test-dev (AP, last in 2024); DOTA (mAP, last in 2022); DOTA test (mAP, last in 2022).
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 · LocNet: Improving Localization Accuracy for Object Detection (CVPR 2016 · 166 citations)
- 2017 · Finding Tiny Faces (CVPR 2017 · 748 citations)
- 2018 · Improving Object Localization With Fitness NMS and Bounded IoU Loss (CVPR 2018 · 241 citations)
- 2019 · Learning RoI Transformer for Oriented Object Detection in Aerial Images (CVPR 2019 · 1,624 citations)
- 2020 · Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression (AAAI 2020 · 4,221 citations)
- 2021 · R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object (AAAI 2021 · 1,151 citations)
- 2022 · Oriented RepPoints for Aerial Object Detection (CVPR 2022 · 545 citations)
- 2023 · Phase-Shifting Coder: Predicting Accurate Orientation in Oriented Object Detection (CVPR 2023 · 181 citations)
- 2024 · Rethinking Boundary Discontinuity Problem for Oriented Object Detection (CVPR 2024 · 42 citations)
- 2025 · FBRT-YOLO: Faster and Better for Real-Time Aerial Image Detection (AAAI 2025 · 157 citations)
- 2026 · Generalized-Scale Object Counting with Gradual Query Aggregation (AAAI 2026 · 2 citations)
Earliest papers here
- LocNet: Improving Localization Accuracy for Object Detection (CVPR 2016 · 166 citations)
- Submodular Asymmetric Feature Selection in Cascade Object Detection (AAAI 2016 · 11 citations)
- Finding Tiny Faces (CVPR 2017 · 748 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 minival · AP (higher is better)
8 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 41.6 | Feature Selective Anchor-Free Module for Single-Shot Object Detection (CVPR 2019 · 1,059 citations) |
| 2020 | 42.3 | Learning From Noisy Anchors for One-Stage Object Detection (CVPR 2020 · 116 citations) |
COCO test-dev · AP (higher is better)
8 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 44.6 | Feature Selective Anchor-Free Module for Single-Shot Object Detection (CVPR 2019 · 1,059 citations) |
| 2020 | 48.9 | A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection (NeurIPS 2020 · 27 citations) |
| 2021 | 50.6 | Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection (CVPR 2021 · 439 citations) |
| 2022 | 55.5 | Confidence Propagation Cluster: Unleash Full Potential of Object Detectors (CVPR 2022 · 16 citations) |
MS COCO test-dev · AP (higher is better)
6 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 50.7 | Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (CVPR 2020 · 2,370 citations) |
DOTA · mAP (higher is better)
5 papers, 2021–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 76.95 | Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (AAAI 2021 · 361 citations) |
DOTA test · mAP (higher is better)
4 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 69.56 | Learning RoI Transformer for Oriented Object Detection in Aerial Images (CVPR 2019 · 1,624 citations) |
| 2021 | 74.1 | Learning Modulated Loss for Rotated Object Detection (AAAI 2021 · 320 citations) |
| 2022 | 79.16 | Polygon-to-Polygon Distance Loss for Rotated Object Detection (AAAI 2022 · 43 citations) |
What has been tried
81 papers here report 999 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 6%, hurt 19%, 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 |
|---|---|---|---|---|
| reduce inherent network redundancy | VisDrone · AP | 23.6 -> 23.4 | Hurt | FBRT-YOLO: Faster and Better for Real-Time Aerial Image Detection (AAAI 2025 · 157 citations) |
| add GatedFFN | VisDrone val · AP95 | 19.1 -> 20.3; ↑ 1.2 | Helped | RemDet: Rethinking Efficient Model Design for UAV Object Detection (AAAI 2025 · 44 citations) |
| Use a longer encoding length for CSL | HRSC2016 · AP75 | 2.56 -> 13.58 -> 61.43 -> 77.76 | Helped | Rethinking Boundary Discontinuity Problem for Oriented Object Detection (CVPR 2024 · 42 citations) |
| Use IoU scores instead of one-hot coding for OBB classification | DOTA-v1.0 · mAP50 | 73.46 -> 74.00 | Helped | Theoretically Achieving Continuous Representation of Oriented Bounding Boxes (CVPR 2024 · 31 citations) |
| Add CFC | COD10K · Fβ | 0.685 -> 0.747 | Helped | Strategic Preys Make Acute Predators: Enhancing Camouflaged Object Detectors by Generating Camouflaged Objects (ICLR 2024 · 29 citations) |
| Add group pooling to the ablation configuration without equivariance transformations | DOTA-v1.0 · mAP | 73.32 -> 73.91 | Helped | FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 2024 · 26 citations) |
| add rotation-equivariant diffusion model (over DiffusionDet baseline) | DIOR-R · AP50 | +2.24 (49.91 to 52.15) | Helped | ReDiffDet: Rotation-equivariant Diffusion Model for Oriented Object Detection (CVPR 2025 · 18 citations) |
| Add a semantic feature map to SRA mask regression | MS COCO val2017 · AP | 36.4 -> 38.2 | Helped | Semantic-Aware Transformation-Invariant RoI Align (AAAI 2024 · 3 citations) |
| integrate restoration learning and parameter-space smoothing into YOLOv10 | RTTS · mAP | +3.2 | Helped | Delving into Cascaded Instability: A Lipschitz Continuity View on Image Restoration and Object Detection Synergy (NeurIPS 2025 · 2 citations) |
| vary M from 200 to 300 and N from 100 to 180 | YUD · AA@3° | 70.6 to 78.4; mean 75.3; STDEV 2.2 | No clear effect | DiskVPS: Vanishing Point Detector via Hough Transform in a Disk Region (CVPR 2025 · 0 citations) |
| replace strict rotation-equivariant filters with relaxed rotation-equivariant filters | VOC test, C4 · AP50:95 | 64.2 -> 65.1 (+0.9) | Helped | R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · 0 citations) |
| add prediction score to BoS | Vehicle detection, 10 unseen datasets · RMSE (%) | 2.25 ±0.13 -> 2.24 ±0.12 | No clear effect | Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · 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 |
|---|---|---|---|---|
| ATSS | COCO test-dev | AP | 43.6 – 47.7 | 2 papers, e.g. CVPR 2021 and NeurIPS 2020 |
| Focal Loss | COCO minival | AP | 33.9 – 36.1 | 2 papers, e.g. CVPR 2019 and CVPR 2020 |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- VisDrone · 2 of 7 papers
- UAVDT · 2 of 7 papers
- DOTA v1.0 · 2 of 7 papers
- DOTA v1.5 · 2 of 7 papers
- COCO · 2 of 7 papers
Baselines
- YOLOv8 · 4 of 7 papers
- YOLOv10 · 4 of 7 papers
- YOLOv9 · 2 of 7 papers
- YOLOv5 · 2 of 7 papers
- RT-DETR · 2 of 7 papers
- QueryDet · 2 of 7 papers
- CEASC · 2 of 7 papers
- YOLC · 2 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 36 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Xidian University · 2 papers
- Chinese Academy of Sciences · 2 papers
- Hangzhou Dianzi University · 2 papers
- Shanghai Jiao Tong University · 2 papers
- Alibaba Group · 2 papers
- ETH Zurich · 2 papers
- Australian National 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
- Region-based and query-based detectors (183 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).