Real-Time Object Detection (YOLO and DETR Lines)
Detectors that find and box objects in images fast enough for live camera feeds on one GPU or an edge device.
Tier Grades
Tier A: Frontier AI labs
feasible
Frontier labs and big tech release the reference detectors, for example RT-DETR at 108 FPS on a T4.
Tier B: Indonesian AI labs
feasible
Fine-tuning a detector that runs at 74 to 108 FPS on a T4 GPU on local data is within reach of an Indonesian lab.
Tier C: Institution units such as ITB
feasible
A university unit can train and benchmark these detectors on COCO-scale data with a few GPUs, as the 8-author RT-DETR paper shows.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Pretrained real-time detectors fine-tuned on client images fit a proof of concept, with T4-class inference stated in the anchor.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeRT-DETR-R50 and R101 run at 108 and 74 FPS on a T4 GPU; training compute is not stated in the abstract.
- DataCOCO, with Objects365 pretraining lifting RT-DETR-R50 and R101 to 55.3% and 56.2% AP.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 8 authors on the RT-DETR paper (arXiv 2304.08069).
- Time horizonMonths: new YOLO and DETR variants appear several times a year.
- Shelf-life signalsA new YOLO or DETR release overtakes the speed and accuracy curve on COCO, Open-vocabulary detectors reach the same speed and absorb closed-set training
- Pipeline roleTier A and big tech release real-time detectors; Tier C benchmarks and adapts them to local datasets; Tier D fine-tunes them for client proofs of value such as counting and inspection; Tier B packages local datasets.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Baidu
Peking University
Roboflow
Ultralytics
Tsinghua University
University of Science and Technology of China (USTC)
Institute of Artificial Intelligence, Hefei Comprehensive National Science CenterLinked models
YOLO26visionexemplarRF-DETRvisionexemplarD-FINEvisionexemplarYOLOv10visionexemplarRT-DETRvisionexemplarYOLO26lvisionproduced byYOLO26mvisionproduced byYOLO26nvisionproduced byYOLO26svisionproduced byYOLO26xvisionproduced byRF-DETR 2XLvisionproduced byRF-DETR Lvisionproduced byRF-DETR Mvisionproduced byRF-DETR Nvisionproduced byRF-DETR Svisionproduced byRF-DETR XLvisionproduced by
YOLO12visionproduced by
DEIMvisionproduced byYOLO11visionproduced by
YOLOv9visionproduced by
DINOv2visionused by
YOLO-NASvisionproduced by
DDQvisionproduced by
RTMDetvisionproduced by+88
DAMO-YOLOvisionproduced by
Co-DETRvisionproduced byDiffusionDetvisionproduced by
YOLOv6visionproduced byYOLOv7visionproduced byViTDetvisionproduced byPP-YOLOEvisionproduced by
DINO (DETR with improved denoising anchor boxes)visionproduced byDINO detectorvisionproduced byDN-DETRvisionproduced byDAB-DETRvisionproduced by
TOODvisionproduced by
Conditional DETRvisionproduced by
YOLOXvisionproduced by
YOLOSvisionproduced byYOLORvisionproduced by
YOLOFvisionproduced byD2Govisionproduced bySparse R-CNNvisionproduced byDeformable DETRvisionproduced by
DetectoRSvisionproduced byDETRvisionproduced byDETR ResNet-50visionproduced byDETR ResNet-101visionproduced byDETR ResNet-50 panopticvisionproduced byYOLOv4visionproduced by
CentripetalNetvisionproduced by
SpineNetvisionproduced byEfficientDetvisionproduced byEfficientDet-D0visionproduced byEfficientDet-D7visionproduced byEfficientDet-Litevisionproduced byDetectron2 model zoovisionproduced by
SELSAvisionproduced by
NAS-FCOSvisionproduced byRepPointsvisionproduced byNAS-FPNvisionproduced by
CenterNet (Objects as Points)visionproduced byFoveaBoxvisionproduced by
FCOSvisionproduced by
NVIDIA SSD300visionproduced by
TridentNetvisionproduced byGrid R-CNNvisionproduced by
CornerNetvisionproduced by
YOLOv3visionproduced bySSD MobileNet v2visionproduced bySSDLitevisionproduced by
Cascade R-CNNvisionproduced byLSTM mobile video object detectorvisionproduced byRetinaNetvisionproduced byFGFAvisionproduced byYOLOv2 (YOLO9000)visionproduced byDFF (Deep Feature Flow)visionproduced by
SSD (Single Shot MultiBox Detector)visionproduced bySSD-ResNet34 (MLPerf)visionproduced byYOLO (original line, v1 to v3)visionproduced byYOLOv1visionproduced by
Faster R-CNNvisionproduced byFast R-CNNvisionproduced by
R-CNN familyvisionproduced byR-CNNvisionproduced byLW-DETRvisionproduced byYOLO27visionproduced byYOLOv11 Largevisionproduced byYOLOv5visionproduced byYOLOv8visionproduced byR-FCNvisionproduced byExtremeNetvisionproduced byPix2Seqvisionproduced byNanoDetvisionproduced byFastestDetvisionproduced byCAEv2visionused byYOLOX-Posevisionproduced byPolyp Detection RT-DETR (Holoscan)visionproduced byNVIDIA TAO pretrained detection and segmentation modelsvisionproduced byMobile Object Localizervisionproduced byCPPE-5 medical PPE detectorsvisionproduced byRoboflow 3.0visionproduced byRoboflow Instantvisionproduced byYOLOlitevisionproduced by
YOLO-Provisionproduced byFOMO (Faster Objects, More Objects)TinyMLproduced bySSD MobileNet v1TinyMLproduced by
ST YOLO detectorsTinyMLproduced byST YOLO LC v1TinyMLproduced byST YOLOD v2 (Tiny, Milli)TinyMLproduced byST YOLOX-nanoTinyMLproduced byYOLO v3 TinyTinyMLproduced byLinks
R&D TopicReal-Time Object Detection (YOLO and DETR Lines)
ModelsRT-DETRBaidu, visionRF-DETRRoboflow, visionYOLOv10Tsinghua University, visionYOLO26Ultralytics, vision+108 more
RT-DETRBaidu, visionRF-DETRRoboflow, visionYOLOv10Tsinghua University, visionYOLO26Ultralytics, vision+108 more
Models 112
YOLO26Ultralytics, visionexemplarRF-DETRRoboflow, visionexemplarR-CNN familyUC Berkeley, visionproduced bySSD (Single Shot MultiBox Detector)University of North Carolina at Chapel Hill, Google, visionproduced byDEIMIntellindust AI Lab, City University of Hong Kong, Great Bay University, Hefei Normal University, visionproduced byDINO (DETR with improved denoising anchor boxes)IDEA Research (International Digital Economy Academy), visionproduced byDINOv2Meta, visionused byST YOLO detectorsSTMicroelectronics, TinyMLproduced byYOLO (original line, v1 to v3)University of Washington, Allen AI, visionproduced byYOLOv2 (YOLO9000)University of Washington, Allen AI, visionproduced byD-FINEUniversity of Science and Technology of China (USTC), Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, visionexemplarDETRMeta, visionproduced by+100
EfficientDetGoogle, visionproduced byEfficientDet-LiteGoogle, visionproduced byYOLOv1University of Washington, Allen AI, visionproduced byDETR ResNet-50 panopticMeta, visionproduced byDFF (Deep Feature Flow)University of Science and Technology of China (USTC), Microsoft Research, visionproduced byFaster R-CNNMicrosoft Research, University of Science and Technology of China (USTC), visionproduced byFGFAUniversity of Science and Technology of China (USTC), Microsoft Research, visionproduced byRT-DETRBaidu, Peking University, visionexemplarSSD MobileNet v2Google, visionproduced bySSD-ResNet34 (MLPerf)University of North Carolina at Chapel Hill, Google, visionproduced byST YOLO LC v1STMicroelectronics, TinyMLproduced byST YOLOD v2 (Tiny, Milli)STMicroelectronics, TinyMLproduced byST YOLOX-nanoSTMicroelectronics, TinyMLproduced byYOLO11Ultralytics, visionproduced byYOLOv3University of Washington, visionproduced byYOLOv8Ultralytics, visionproduced byCascade R-CNNUC San Diego, visionproduced byCenterNet (Objects as Points)University of Texas at Austin, UC Berkeley, visionproduced byCentripetalNetXi'an Jiaotong University, SenseTime, visionproduced byDAB-DETRIDEA Research (International Digital Economy Academy), visionproduced byDDQThe University of Hong Kong, Shanghai AI Lab, visionproduced byDeformable DETRSenseTime, University of Science and Technology of China (USTC), visionproduced byDetectoRSJohns Hopkins University, Google, visionproduced byDETR ResNet-101Meta, visionproduced byDETR ResNet-50Meta, visionproduced byDiffusionDetThe University of Hong Kong, Tencent, visionproduced byDINO detectorIDEA Research (International Digital Economy Academy), visionproduced byDN-DETRIDEA Research (International Digital Economy Academy), visionproduced byEfficientDet-D0Google, visionproduced byEfficientDet-D7Google, visionproduced byFast R-CNNMicrosoft Research, visionproduced byFOMO (Faster Objects, More Objects)Edge Impulse, TinyMLproduced byLSTM mobile video object detectorGoogle, visionproduced byMobile Object LocalizerGoogle, visionproduced byNAS-FCOSNorthwestern Polytechnical University, University of Adelaide, visionproduced byNVIDIA SSD300NVIDIA, visionproduced byNVIDIA TAO pretrained detection and segmentation modelsNVIDIA, visionproduced byPolyp Detection RT-DETR (Holoscan)NVIDIA, visionproduced byR-CNNUC Berkeley, visionproduced byRepPointsPeking University, Microsoft Research, visionproduced byRF-DETR 2XLRoboflow, visionproduced byRF-DETR LRoboflow, visionproduced byRF-DETR MRoboflow, visionproduced byRF-DETR NRoboflow, visionproduced byRF-DETR SRoboflow, visionproduced byRF-DETR XLRoboflow, visionproduced bySELSAMcGill University, TuSimple, visionproduced bySparse R-CNNThe University of Hong Kong, ByteDance, visionproduced bySSDLiteGoogle, visionproduced byTridentNetUniversity of Chinese Academy of Sciences, TuSimple, visionproduced byYOLO-ProEdge Impulse, visionproduced byYOLO12University at Buffalo, University of Chinese Academy of Sciences, visionproduced byYOLO26lUltralytics, visionproduced byYOLO26mUltralytics, visionproduced byYOLO26nUltralytics, visionproduced byYOLO26sUltralytics, visionproduced byYOLO26xUltralytics, visionproduced byYOLO27Ultralytics, visionproduced byYOLOFInstitute of Automation, Chinese Academy of Sciences, Megvii, visionproduced byYOLOliteRoboflow, visionproduced byYOLOv10Tsinghua University, visionexemplarYOLOX-PoseOpenMMLab, visionproduced byCAEv2visionused byCo-DETRSenseTime, visionproduced byConditional DETRMicrosoft AI, visionproduced byCornerNetPrinceton University, visionproduced byCPPE-5 medical PPE detectorsvisionproduced byD2GoMeta, visionproduced byDAMO-YOLOAlibaba DAMO, visionproduced byDetectron2 model zooMeta, visionproduced byExtremeNetUniversity of Texas at Austin, visionproduced byFCOSUniversity of Adelaide, visionproduced byGrid R-CNNSenseTime, visionproduced byLW-DETRvisionproduced byNAS-FPNGoogle, visionproduced byPix2SeqGoogle, visionproduced byPP-YOLOEBaidu, visionproduced byR-FCNMicrosoft Research, visionproduced byRetinaNetMeta, visionproduced byRoboflow 3.0Roboflow, visionproduced byRoboflow InstantRoboflow, visionproduced byRTMDetOpenMMLab, visionproduced bySpineNetGoogle, visionproduced bySSD MobileNet v1TinyMLproduced byTOODIntellifusion, visionproduced byViTDetMeta, visionproduced byYOLO v3 TinyTinyMLproduced byYOLO-NASDeci, visionproduced byYOLORAcademia Sinica, visionproduced byYOLOSHuazhong University of Science and Technology, visionproduced byYOLOv11 LargeUltralytics, visionproduced byYOLOv4Academia Sinica, visionproduced byYOLOv5Ultralytics, visionproduced byYOLOv6Meituan, visionproduced byYOLOv7Academia Sinica, visionproduced byYOLOv9Academia Sinica, visionproduced byYOLOXMegvii, visionproduced byFastestDetvisionproduced byFoveaBoxvisionproduced byNanoDetvisionproduced byPapers
- You Only Look Once: Unified, Real-Time Object DetectionJoseph Redmon et al., arXiv preprint, 2015 founding
- YOLOv10: Real-Time End-to-End Object DetectionAo Wang et al., NeurIPS 2024 (per arXiv comment), 2024 state of the art
- DETRs Beat YOLOs on Real-time Object DetectionYian Zhao et al., arXiv preprint, 2023 state of the art
- End-to-End Object Detection with TransformersNicolas Carion et al., arXiv preprint, 2020 state of the art
- A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NASJuan Terven et al., Mach. Learn. Knowl. Extr. 2023, 5, 1680-1716, 2023 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- DETRs Beat YOLOs on Real-time Object Detection arxiv.org https://arxiv.org/abs/2304.08069
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Real-Time Object Detection (YOLO and DETR Lines)", https://aurion.semai.id/topics/real-time-object-detection/, records of October 2026.