An event-camera system with active gaze control and contrast-maximization spin estimation achieves real-time performance in table tennis with 8.8% magnitude error, 6.4° axis error, 3 ms latency, and 750 Hz throughput.
hub Mixed citations
Yolov9: Learning what you want to learn using programmable gradient information
Mixed citation behavior. Most common role is background (60%).
21
Pith papers citing it
352
external citations · Pith
Background
60% of classified citations
abstract
Today's deep learning methods focus on how to design the most appropriate objective functions so that the prediction results of the model can be closest to the ground truth. Meanwhile, an appropriate architecture that can facilitate acquisition of enough information for prediction has to be designed. Existing methods ignore a fact that when input data undergoes layer-by-layer feature extraction and spatial transformation, large amount of information will be lost. This paper will delve into the important issues of data loss when data is transmitted through deep networks, namely information bottleneck and reversible functions. We proposed the concept of programmable gradient information (PGI) to cope with the various changes required by deep networks to achieve multiple objectives. PGI can provide complete input information for the target task to calculate objective function, so that reliable gradient information can be obtained to update network weights. In addition, a new lightweight network architecture -- Generalized Efficient Layer Aggregation Network (GELAN), based on gradient path planning is designed. GELAN's architecture confirms that PGI has gained superior results on lightweight models. We verified the proposed GELAN and PGI on MS COCO dataset based object detection. The results show that GELAN only uses conventional convolution operators to achieve better parameter utilization than the state-of-the-art methods developed based on depth-wise convolution. PGI can be used for variety of models from lightweight to large. It can be used to obtain complete information, so that train-from-scratch models can achieve better results than state-of-the-art models pre-trained using large datasets, the comparison results are shown in Figure 1. The source codes are at: https://github.com/WongKinYiu/yolov9.
hub tools
citation-role summary
background 3
method 2
citation-polarity summary
representative citing papers
DHNet with patch alignment and dual hypergraph fusion reaches SOTA RGBT video object detection on VT-VOD50 and the new large-scale DVT-VOD1000 benchmark.
Presents MMIO benchmark and RTVP method achieving state-of-the-art 42.2% AP in zero-shot industrial defect detection.
RefDiffNet is a lightweight input enhancement block that uses reference image comparison to expose PCB defects, delivering up to 18% relative mAP50:95 gains across YOLO, RT-DETR, and Faster R-CNN detectors with 0.004-0.005M extra parameters.
Releases a recycling-specific dataset of >10k images and evaluates YOLO variants on small dense overlapping objects with augmentation and anomaly detection.
A single network achieves any-depth object detection by splitting stages into always-executed essential paths and skippable refinement paths, trained via self-distillation on the full and minimal extremes to maintain stage compatibility.
SPAR is a street-legal physical rim that cuts modern ALPR accuracy by 60% and reaches 18% targeted impersonation while costing under $100 and requiring no plate modification.
YOLOv12 is a new attention-based real-time object detector that reports higher accuracy than YOLOv10, YOLOv11, and RT-DETR variants at comparable or better speed and efficiency.
Hippocampus-DETR integrates a hippocampal memory network (HipNet) into DETR to simulate brain subregions for pattern separation, completion, and improved detection accuracy plus generalization.
MV3DT is a fully distributed real-time multi-view 3D tracking framework achieving 94.3% IDF1 on WILDTRACK while scaling to 100 cameras at 30 FPS with under 10 ms latency and 2.2% communication overhead in a zero-shot regime.
Multi-view fusion from three satellite viewpoints boosts mAP50 by up to 36.3% and mAP50-95 by 46.5% over single-view baselines in YOLO detectors for space object detection.
A drone-mounted stereo camera pipeline with YOLO segmentation, deep stereo depth, centroid triangulation, and MAD outlier rejection achieves robust 3D positioning of thin pine branches at 1-2 m distances.
A scale-robust lightweight CNN for glottis segmentation achieves 92.9% mDice at over 170 FPS with a 19 MB model size on three datasets.
A unified pipeline using OCR, inpainting, and diffusion models restores text in degraded documents on a new synthetic benchmark dataset, evaluated with the proposed UCSM metric.
Introduces UAVDB dataset for UAV detection/segmentation via PIC point-to-box conversion and SAM2 masks, with YOLO baselines showing PIC+SAM2 outperforms prior annotation methods on IoU.
A hierarchically decoupled heterogeneous MoE framework with YOLO experts and lightweight gating network reports 76.8% mAP50-95 on a composite traffic sign dataset, a 2.3% gain over baseline with 39.4% lower compute.
DFIR-DETR augments RT-DETR with frequency-domain iterative refinement and dynamic feature aggregation, reporting 92.9% mAP50 on NEU-DET and 51.6% on VisDrone at 11.7M parameters and 47.2 GFLOPs.
A deep learning pipeline with YOLOv8 and Keypoint R-CNN achieves ICC up to 0.80 for bone loss severity and 87% accuracy for horizontal vs. angular pattern classification on 1000 annotated IOPA radiographs.
Drone stereo vision pipeline segments pine branches with YOLO variants and estimates depth with deep stereo networks, yielding more coherent maps than SGBM at 1-2 m distances.
Comparative review of YOLOv8 to YOLO11 architectures based on papers, docs, and code inspection, noting incremental improvements and some unchanged blocks.
YOLOv11 adds blocks such as C3k2, SPPF, and C2PSA to improve feature extraction, mAP, and efficiency while supporting detection, segmentation, pose, and oriented detection across model sizes.
citing papers explorer
-
Event-based Gaze Control System for Accurate Real-time Spin Estimation in Professional Ball Games
An event-camera system with active gaze control and contrast-maximization spin estimation achieves real-time performance in table tennis with 8.8% magnitude error, 6.4° axis error, 3 ms latency, and 750 Hz throughput.
-
Dual-Correlation Hypergraph Network for Unaligned RGBT Video Object Detection and A Large-scale Benchmark
DHNet with patch alignment and dual hypergraph fusion reaches SOTA RGBT video object detection on VT-VOD50 and the new large-scale DVT-VOD1000 benchmark.
-
Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline
Presents MMIO benchmark and RTVP method achieving state-of-the-art 42.2% AP in zero-shot industrial defect detection.
-
RefDiffNet: Learning to Expose Subtle PCB Defects Before Detection
RefDiffNet is a lightweight input enhancement block that uses reference image comparison to expose PCB defects, delivering up to 18% relative mAP50:95 gains across YOLO, RT-DETR, and Faster R-CNN detectors with 0.004-0.005M extra parameters.
-
Small Object Detection in Industrial Recycling: A New Dataset and YOLO Performance Evaluation
Releases a recycling-specific dataset of >10k images and evaluates YOLO variants on small dense overlapping objects with augmentation and anomaly detection.
-
AnyDepth-DETR/-YOLO: Any-depth object detection with a single network
A single network achieves any-depth object detection by splitting stages into always-executed essential paths and skippable refinement paths, trained via self-distillation on the full and minimal extremes to maintain stage compatibility.
-
Street-Legal Physical-World Adversarial Rim for License Plates
SPAR is a street-legal physical rim that cuts modern ALPR accuracy by 60% and reaches 18% targeted impersonation while costing under $100 and requiring no plate modification.
-
YOLOv12: Attention-Centric Real-Time Object Detectors
YOLOv12 is a new attention-based real-time object detector that reports higher accuracy than YOLOv10, YOLOv11, and RT-DETR variants at comparable or better speed and efficiency.
-
Hippocampus-DETR: An Explicit Memory Object Detection Framework Based on Hippocampus Modeling
Hippocampus-DETR integrates a hippocampal memory network (HipNet) into DETR to simulate brain subregions for pattern separation, completion, and improved detection accuracy plus generalization.
-
Fully Distributed Multi-View 3D Tracking in Real-Time
MV3DT is a fully distributed real-time multi-view 3D tracking framework achieving 94.3% IDF1 on WILDTRACK while scaling to 100 cameras at 30 FPS with under 10 ms latency and 2.2% communication overhead in a zero-shot regime.
-
Collaborative Space Object Detection with Multi-Satellite Viewpoints in LEO Constellations
Multi-view fusion from three satellite viewpoints boosts mAP50 by up to 36.3% and mAP50-95 by 46.5% over single-view baselines in YOLO detectors for space object detection.
-
Low-Cost Stereo Vision for Robust 3D Positioning of Thin Radiata Pine Branches in Autonomous Drone Pruning
A drone-mounted stereo camera pipeline with YOLO segmentation, deep stereo depth, centroid triangulation, and MAD outlier rejection achieves robust 3D positioning of thin pine branches at 1-2 m distances.
-
A Real-time Scale-robust Network for Glottis Segmentation in Nasal Transnasal Intubation
A scale-robust lightweight CNN for glottis segmentation achieves 92.9% mDice at over 170 FPS with a 19 MB model size on three datasets.
-
DocRevive: A Unified Pipeline for Document Text Restoration
A unified pipeline using OCR, inpainting, and diffusion models restores text in degraded documents on a new synthetic benchmark dataset, evaluated with the proposed UCSM metric.
-
UAVDB: Point-Guided Masks for UAV Detection and Segmentation
Introduces UAVDB dataset for UAV detection/segmentation via PIC point-to-box conversion and SAM2 masks, with YOLO baselines showing PIC+SAM2 outperforms prior annotation methods on IoU.
-
Hierarchically Decoupled Mixture-of-Experts for Robust Traffic Sign Recognition in Complex Driving Scenarios
A hierarchically decoupled heterogeneous MoE framework with YOLO experts and lightweight gating network reports 76.8% mAP50-95 on a composite traffic sign dataset, a 2.3% gain over baseline with 39.4% lower compute.
-
DFIR-DETR: Frequency-Domain Iterative Refinement and Dynamic Feature Aggregation for Small Object Detection
DFIR-DETR augments RT-DETR with frequency-domain iterative refinement and dynamic feature aggregation, reporting 92.9% mAP50 on NEU-DET and 51.6% on VisDrone at 11.7M parameters and 47.2 GFLOPs.
-
AI-assisted radiographic analysis in detecting alveolar bone-loss severity and patterns
A deep learning pipeline with YOLOv8 and Keypoint R-CNN achieves ICC up to 0.80 for bone loss severity and 87% accuracy for horizontal vs. angular pattern classification on 1000 annotated IOPA radiographs.
-
Positioning radiata pine branches requiring pruning by drone stereo vision
Drone stereo vision pipeline segments pine branches with YOLO variants and estimates depth with deep stereo networks, yielding more coherent maps than SGBM at 1-2 m distances.
-
YOLOv8 to YOLO11: A Comprehensive Architecture In-depth Comparative Review
Comparative review of YOLOv8 to YOLO11 architectures based on papers, docs, and code inspection, noting incremental improvements and some unchanged blocks.
-
YOLOv11: An Overview of the Key Architectural Enhancements
YOLOv11 adds blocks such as C3k2, SPPF, and C2PSA to improve feature extraction, mAP, and efficiency while supporting detection, segmentation, pose, and oriented detection across model sizes.