场景适配方案
针对零件表面划痕、焊接缺陷等典型问题:
Implementation Steps:
- 数据准备阶段:
- 采集200+缺陷样本(需包含多种光照条件)
- utilization
labelme
标注缺陷区域(导出JSON格式)
- Model Selection:
- 高精度场景:选用
yoloe-v8l-seg.pt
Pre-trained models - 实时产线:改用
yoloe-s-det.pt
纯检测模型
- 高精度场景:选用
- 部署优化:
- pass (a bill or inspection etc)
export.py
输出ONNX/TensorRT格式 - set up
--conf 0.7
避免误检(质检需严格)
- pass (a bill or inspection etc)
Typical workflow:
# 视觉提示模式(适合外观变异大的缺陷) python predict_visual_prompt.py --source conveyor.jpg --ref defect_sample1.jpg --checkpoint yoloe-l.pt
Caveats:对反光金属件,需先进行cv2.GaussianBlur
preprocessing
This answer comes from the articleYOLOE: an open source tool for real-time video detection and segmentation of objectsThe