ParsaVictor/pcb-bare-board-defect-detection-yolo26

PCB Bare-Board Defect Detection (YOLO26s)

A YOLO26-based object detector that finds manufacturing defects on bare printed circuit boards — the inspection step before any components are placed: missing holes, mouse bites, open circuits, shorts, spurs, spurious copper, and pin holes.

Part of the PCB Inspection Suite. Trained on ParsaVictor/pcb-bare-board-defect-detection.

Metrics

Split	mAP50	mAP50-95
val	0.895	0.601
test	0.928	0.617

Trained: YOLO26s, imgsz 1024, 80 epochs (best at epoch 17), online augmentation, close_mosaic scheduled.

Known weaknesses

• short class recall is weaker (F1-optimal threshold ≈0.41, lower confidence than most classes)
• open_circuit recall drops to ~70% at its F1-optimal threshold (0.77) — prefer the lower global threshold if recall matters more than precision for this class

Per-class confidence thresholds (F1-optimal on val)

Class	Threshold
missing_hole	0.745
mouse_bite	0.484
open_circuit	0.770
short	0.406
spur	0.382
spurious_copper	0.330
pin_hole	0.571

A global fallback threshold of 0.465 is used for any class with too few validation examples to tune individually.

Usage

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

weights = hf_hub_download("ParsaVictor/pcb-bare-board-defect-detection-yolo26", "best.pt")
model = YOLO(weights)

results = model.predict("your_board_photo.jpg", conf=0.465)
results[0].show()

For best accuracy, apply the per-class thresholds above instead of one global confidence value — see thresholds.json in this repo.

Related

• Training dataset: pcb-bare-board-defect-detection
• Sibling model: pcb-solder-defect-detection-yolo26
• Full suite overview: pcb-inspection-suite

License

Research/development use — see the dataset license. Contact the author before commercial use.