Steel Surface Defect Detection — ResNet-50
Author: Md. Sajjad Ullah
Results
| Model | Accuracy | F1-Score |
|---|---|---|
| ResNet-50 | 100.00% | 100.00% |
| EfficientNet-B3 | 98.89% | 98.89% |
| MobileNet-V3 | 98.52% | 98.52% |
| Custom CNN | 85.19% | 84.95% |
Dataset
NEU Surface Defect Database — 1,800 images | 6 classes
Crazing · Inclusion · Patches · Pitted Surface · Rolled-in Scale · Scratches
How to Use
import torch
from torchvision.models import resnet50
model = resnet50()
model.fc = torch.nn.Sequential(
torch.nn.Dropout(0.4),
torch.nn.Linear(2048, 6)
)
model.load_state_dict(
torch.load("best_ResNet-50.pth", map_location="cpu")
)
model.eval()