Pith. sign in

REVIEW 18 cited by

RTMDet: An Empirical Study of Designing Real-Time Object Detectors

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2212.07784 v2 pith:6WJ7WQWE submitted 2022-12-14 cs.CV

Chengqi Lyu , Wenwei Zhang , Haian Huang , Yue Zhou , Yudong Wang , Yanyi Liu , Shilong Zhang , Kai Chen This is my paper · ORCID
classification cs.CV
keywords objectreal-timertmdetdetectorsachievesarchitecturedesigningdetection
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

In this paper, we aim to design an efficient real-time object detector that exceeds the YOLO series and is easily extensible for many object recognition tasks such as instance segmentation and rotated object detection. To obtain a more efficient model architecture, we explore an architecture that has compatible capacities in the backbone and neck, constructed by a basic building block that consists of large-kernel depth-wise convolutions. We further introduce soft labels when calculating matching costs in the dynamic label assignment to improve accuracy. Together with better training techniques, the resulting object detector, named RTMDet, achieves 52.8% AP on COCO with 300+ FPS on an NVIDIA 3090 GPU, outperforming the current mainstream industrial detectors. RTMDet achieves the best parameter-accuracy trade-off with tiny/small/medium/large/extra-large model sizes for various application scenarios, and obtains new state-of-the-art performance on real-time instance segmentation and rotated object detection. We hope the experimental results can provide new insights into designing versatile real-time object detectors for many object recognition tasks. Code and models are released at https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 18 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LEVIRDet: A Million-Scale 159-Category Dataset and Foundation Model for Universal Remote Sensing Object Detection

    cs.CV 2026-06 unverdicted novelty 6.5 of 10

    LEVIRDet-159 is a 159-category remote sensing detection dataset with 2.56M boxes exceeding prior scales; LEVIRDetNet achieves SOTA zero-shot performance on 9 external benchmarks with 5.02 mAP average improvement.

  2. What to Distinguish and How? Opportunities and Challenges of Augmenting Multiple, Cluttered Objects in Complex Scenes for People with Low Vision

    cs.HC 2026-07 conditional novelty 6.0 of 10

    For people with low vision, AR overlays that rank objects by importance redirect attention toward high-priority objects, but multi-object augmentation lowers overall scene recall and creates new visual-confusion problems.

  3. Fully Rotation-Equivariant Spectral-Spatial Learning for Multispectral Object Detection

    cs.CV 2026-07 conditional novelty 6.0 of 10

    FressDet models multispectral channels as a continuous coordinate-warped field inside a C4-rotation-equivariant detector, reporting state-of-the-art mAP on five benchmarks with 2.3M parameters.

  4. KD-Judge: A Knowledge-Driven Automated Judge Framework for Functional Fitness Movements on Edge Devices

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    KD-Judge structures fitness rules via LLM retrieval and chain-of-thought, then uses pose-guided kinematics for rule-based rep validation with caching for efficient edge deployment, achieving RTF < 1 and speedups up to...

  5. RiO-DETR: DETR for Real-time Oriented Object Detection

    cs.CV 2026-03 conditional novelty 6.0 of 10

    RiO-DETR gives the first real-time oriented DETR, matching or beating CNN real-time detectors on DOTA-1.0, DIOR-R, and FAIR-1M-2.0 with a new speed-accuracy trade-off.

  6. MSMVD: Exploiting Multi-scale Image Features via Multi-scale BEV Features for Multi-view Pedestrian Detection

    cs.CV 2025-08 conditional novelty 6.0 of 10

    MSMVD projects multi-scale image features from each camera into multi-scale bird's-eye-view features, then fuses them with a feature pyramid, reaching state-of-the-art MODA on GMVD, Wildtrack, and MultiviewX.

  7. Image Quality Assessment for Machines: Paradigm, Large-scale Database, and Models

    cs.CV 2025-08 conditional novelty 6.0 of 10

    A 2.5-million-image machine-centric quality database and a region-aware model show that human-perception metrics poorly predict machine vision performance under degradations.

  8. A real-time RGB-D perception pipeline for autonomous impact hammers in mining: self-filtering, rock segmentation and rock-breaking poses generation

    cs.RO 2026-07 conditional novelty 5.0 of 10

    An RGB-D perception pipeline that runs at about 10 Hz on embedded hardware segmented rocks and generated feasible rock-breaking poses with a 72% success rate in scaled real-world trials.

  9. Exploring Light-Weight Object Recognition for Real-Time Document Detection

    cs.CV 2025-09 conditional novelty 5.0 of 10

    Adapting IWPOD-Net to document detection gives a 1.8M-parameter rectifier that is faster than YOLO11, RTMDet, and Jdeskew while keeping OCR quality competitive on a synthetic ID dataset.

  10. TRUDI and TITUS: A Multi-Perspective Dataset and A Three-Stage Recognition System for Transportation Unit Identification

    cs.CV 2025-08 conditional novelty 5.0 of 10

    TRUDI is a new public benchmark of 35,034 labeled port images from aerial and ground views, with a three-stage TITUS pipeline that reads IDs but achieves only 12-22 percent end-to-end accuracy.

  11. Measuring the Impact of Rotation Equivariance on Aerial Object Detection

    cs.CV 2025-07 conditional novelty 5.0 of 10

    MessDet shows that strict rotation equivariance in backbone and neck improves aerial detection accuracy over approximate equivariance, achieving SOTA on DOTA-v1.0/v1.5 and DIOR-R with 18.1M parameters.

  12. VisAlgae 2023: A Dataset and Challenge for Algae Detection in Microscopy Images

    cs.CV 2025-05 conditional novelty 5.0 of 10

    The VisAlgae 2023 dataset and challenge provide a new public benchmark for detecting six microalgae species in microscopy images, with baseline and top-10 leaderboard results.

  13. SemSegBench & DetecBench: Benchmarking Reliability and Generalization Beyond Classification

    cs.CV 2025-05 conditional novelty 5.0 of 10

    A large-scale benchmark of 76 segmentation and 61 detection models shows that robustness to attacks and corruptions does not reliably track clean accuracy, and that transformer backbones generalize better under shift.

  14. MHAF-YOLO: Multi-Branch Heterogeneous Auxiliary Fusion YOLO for accurate object detection

    cs.CV 2025-02 conditional novelty 5.0 of 10

    MHAF-YOLO reaches 48.9% AP on COCO with 7.1M parameters by combining auxiliary shallow-deep fusion paths and heterogeneous reparameterized convolutions.

  15. EFSI-DETR: Efficient Frequency-Semantic Integration for Real-Time Small Object Detection in UAV Imagery

    cs.CV 2026-01 conditional novelty 4.0 of 10

    EFSI-DETR, an RT-DETR variant with frequency-flavored fusion, a semantic concentrator, and shallow-feature retention, reports 33.1 AP and 24.8 APs on VisDrone at 188 FPS — a claimed +1.6 AP and +5.8 APs over prior SOTA.

  16. OD-VIRAT: A Large-Scale Benchmark for Object Detection in Realistic Surveillance Environments

    cs.CV 2025-07 conditional novelty 4.0 of 10

    The authors convert VIRAT surveillance videos into COCO-style detection benchmarks and benchmark five detectors on the smaller variant.

  17. CSDN: A Context-Gated Self-Adaptive Detection Network for Real-Time Object Detection

    cs.CV 2025-06 conditional novelty 4.0 of 10

    A plug-and-play Transformer detection head with gated block, neighbor, and deformable attention improves YOLO-family COCO AP by 0.7 to 1.0 points.

  18. O2Former:Direction-Aware and Multi-Scale Query Enhancement for SAR Ship Instance Segmentation

    cs.CV 2025-06 conditional novelty 4.0 of 10

    O2Former adds a multi-scale query generator and an orientation-aware module to Mask2Former and reports improved SAR ship instance segmentation on SSDD and HRSID.

Pith tools