metadata
license: mit
tags:
- pytorch
- onnx
- object-detection
- yolo
- computer-vision
- pcb-inspection
- industrial
Inspection Engine
Model Description
Inspection Engine is a YOLO11s model fine-tuned for automated PCB (Printed Circuit Board) defect detection. It identifies manufacturing defects such as missing components, solder bridges, lifted pads, and other anomalies in real-time, enabling automated quality control in electronics manufacturing.
Model Architecture
- Base Model: YOLO11s (Small variant — optimized for speed)
- Framework: PyTorch + ONNX (for cross-platform deployment)
- Task: Object Detection (Defect Localization)
- Input: High-resolution PCB images
Training Details
- Dataset: PCB defect inspection dataset with annotated defect regions
- Checkpoint:
inspection_engine_final3(best performing run) - Augmentations: Mosaic, color jitter, random affine transforms
- Export: Exported to ONNX for production deployment
Files
| File | Description |
|---|---|
best.pt |
Best PyTorch model weights |
best.onnx |
ONNX export for cross-platform/production deployment |
Usage
PyTorch (Ultralytics)
from ultralytics import YOLO
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(repo_id='devanshty/Inspection-Engine', filename='best.pt')
model = YOLO(model_path)
results = model('pcb_image.jpg')
results[0].show()
ONNX Runtime
import onnxruntime as ort
import numpy as np
from huggingface_hub import hf_hub_download
onnx_path = hf_hub_download(repo_id='devanshty/Inspection-Engine', filename='best.onnx')
session = ort.InferenceSession(onnx_path)
# Prepare input (1, 3, H, W) float32 normalized
input_name = session.get_inputs()[0].name
outputs = session.run(None, {input_name: np.zeros((1, 3, 640, 640), dtype=np.float32)})
Download & Use
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(repo_id='devanshty/Inspection-Engine', filename='best.pt')