Independent and transparent validation of YOLOv13 performance compared to YOLOv12 and YOLO11 on the COCO dataset, addressing performance discrepancies identified in Ultralytics Issue #21243.
- mAP Discrepancy: Official YOLOv12x = 55.2-55.4 vs YOLOv13 paper = 54.8
- Latency Concerns: YOLOv13 slower than YOLOv8 despite claimed improvements
- Need: Independent validation with standardized methodology
- Dataset: MS COCO 2017 validation set (5K images)
- Metrics: mAP50, mAP75, mAP50-95, inference time, memory usage
- Hardware: AMD Ryzen 9 7945HX + RTX 4060 (standardized)
- Reproducibility: Automated scripts + fixed seeds
| Model | Version | Source | Status |
|---|---|---|---|
| YOLOv13-N/S/L/X | Latest | iMoonLab | 🔄 In Progress |
| YOLOv12-N/S/L/X | Latest | sunsmarterjie | 🔄 In Progress |
| YOLO11-N/S/L/X | Latest | Ultralytics | 🔄 In Progress |
| Model | mAP50 | Recall | Improvement |
|---|---|---|---|
| YOLOv13 + SDPA | 82.9% | 73.5% | Baseline |
| YOLOv12 + SDPA | 76.7% | 66.4% | +6.2% for YOLOv13 |
| Model | mAP50-95 | mAP50 | mAP75 | Latency (ms) | Status |
|---|---|---|---|---|---|
| YOLOv13-N | - | - | - | - | ⏳ Planned |
| YOLOv12-N | - | - | - | - | ⏳ Planned |
| YOLO11-N | - | - | - | - | ⏳ Planned |
Complete results expected: 7-10 days
# Clone repository
git clone https://github.com/kennedy-kitoko/yolov13-validation-study.git
cd yolov13-validation-study
# Setup environment
conda create -n yolov13-val python=3.11
conda activate yolov13-val
pip install -r requirements.txt
# Run validation (after complete setup)
python scripts/run_validation.py --model yolov13n --dataset coco- @kennedy-kitoko - Lead Researcher (Beijing Institute of Technology)
- @zshar7 - Initiator & Collaborator
- Community - Cross-validation welcome
- 🍴 Fork the repository
- 🔧 Create a feature branch
- 📊 Add your validation results
- 📝 Document your methodology
- 🔀 Create a Pull Request
-
HyperACE: Hypergraph-based Adaptive Correlation Enhancement
- Treats pixels as hypergraph vertices
- Learnable hyperedge construction for high-order correlations
- Linear complexity message passing module
-
FullPAD: Full-Pipeline Aggregation-and-Distribution Paradigm
- Three separate tunnels for feature forwarding
- Fine-grained information flow across entire pipeline
- Enhanced gradient propagation
-
DS-based Blocks: Model Lightweighting
- Depthwise separable convolutions (DSConv, DS-Bottleneck, DS-C3k2)
- Preserved receptive field with reduced parameters
- Faster inference without accuracy loss
- Clear Performance Metrics: Definitive mAP comparisons on COCO
- Speed-Accuracy Trade-offs: Comprehensive latency analysis
- Training Stability: Multi-run convergence validation
- Broad Compatibility: Cross-platform deployment testing
- Transparent Benchmarking: Open methodology and reproducible results
- Technical Validation: Independent verification of novel techniques
- Best Practices: Standardized evaluation protocols for future models
This study will provide Ultralytics with the evidence needed for informed decision-making:
✅ If YOLOv13 shows clear gains: Evidence-based integration recommendation
✅ If YOLOv13 shows regression: Clear rejection with documented reasoning
✅ Either way: Community gets honest, transparent validation
- Hardware Specifications: Fully documented and reproducible
- Software Environment: Version-locked dependencies
- Statistical Significance: Multiple runs with confidence intervals
- Comparative Analysis: Head-to-head performance tables
- Raw Results: JSON exports from validation runs
- Methodology: Step-by-step reproduction instructions
- Code Availability: All scripts and configurations public
- Issue Tracking: Problems and solutions documented
- Lead Researcher: kitokokennedy13@gmail.com
- Institution: Beijing Institute of Technology
- Discussions: GitHub Discussions
- Issues: Report Issues
- Objective Evaluation: Results reported regardless of outcomes
- Methodology Transparency: All processes fully documented
- Data Sharing: Raw results available for independent analysis
- Peer Review: Community validation encouraged
- Respectful Discussion: Professional discourse in all interactions
- Evidence-Based: Claims supported by reproducible data
- Open Source: All code and data freely available
- Attribution: Proper credit for all contributors
🎯 Goal: Provide objective data for Ultralytics integration decision
🔬 Approach: Rigorous and transparent scientific validation
🤝 Spirit: Open community collaboration
"This study aims to cut through marketing claims with solid scientific evidence, ensuring the YOLO community makes decisions based on reproducible facts rather than promotional materials."
If you use this validation study in your research, please cite:
@misc{yolov13_validation_2025,
title={YOLOv13 Independent Validation Study: Comprehensive Performance Analysis},
author={Kennedy Kitoko and Contributors},
year={2025},
publisher={GitHub},
url={https://github.com/kennedy-kitoko/yolov13-validation-study},
note={Independent validation of YOLOv13 performance claims}
}Latest Update: Repository initialized with validation framework
Next Milestone: Complete COCO validation results (ETA: 7-10 days)
Community Status: Open for contributions and cross-validation
Follow this repository for real-time updates on validation progress and results.