Scribd
Upload
15 views1 page

Fabric Defect Detection Explanation

The document outlines a deep learning pipeline for fabric defect detection using a pre-trained ResNet-18 model to classify images into defect categories. It details the workflow from data preparation to model evaluation and highlights areas for improvement, such as data augmentation, fine-tuning, and enhanced evaluation metrics. The document also suggests deployment strategies for practical applications in real-time defect detection.

Uploaded by

Saisivaram Annadurai
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
15 views1 page

Fabric Defect Detection Explanation

The document outlines a deep learning pipeline for fabric defect detection using a pre-trained ResNet-18 model to classify images into defect categories. It details the workflow from data preparation to model evaluation and highlights areas for improvement, such as data augmentation, fine-tuning, and enhanced evaluation metrics. The document also suggests deployment strategies for practical applications in real-time defect detection.

Uploaded by

Saisivaram Annadurai
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Fabric Defect Detection - Explanation & Scope of

Improvement

1) What the Code Does


The provided code implements a basic deep learning pipeline for fabric defect detection and
classification. It uses a pre-trained ResNet-18 (transfer learning) to classify images of fabric into
different defect categories such as 'good', 'oil', 'thread', etc.

• Loads dataset from folders (train/test) with images organized by defect type.
• Applies transformations (resize, tensor conversion) to prepare images.
• Uses ResNet-18 pre-trained on ImageNet and replaces the last layer to match defect classes.
• Trains the model for 5 epochs using cross-entropy loss and Adam optimizer.
• Evaluates the trained model on the test set and prints classification accuracy.

2) How It Works
The workflow follows the typical deep learning pipeline:
• **Data Preparation**: Images are resized to 128x128 and converted to tensors for input into
PyTorch.
• **Model Setup**: ResNet-18 is loaded, and its final fully connected layer is replaced with one
matching the number of fabric defect categories.
• **Training Loop**: The model is trained for a few epochs where it predicts outputs, calculates loss,
backpropagates errors, and updates weights.
• **Evaluation**: The model predicts defect categories on unseen test data, and accuracy is
computed.

3) Scope of Improvement
While the code works as a simple prototype, there are several areas where it can be improved for
better performance and practical deployment:
• **Data Augmentation**: Add random rotations, flips, color jittering to make the model robust.
• **More Epochs & Fine-tuning**: Train longer and fine-tune all ResNet layers instead of only the last
one.
• **Learning Rate Scheduling**: Use learning rate schedulers to optimize convergence.
• **Evaluation Metrics**: Add precision, recall, and F1-score to understand performance beyond
accuracy.
• **Grad-CAM / Visualization**: Visualize what parts of fabric the model focuses on to ensure
explainability.
• **Deployment**: Convert model to ONNX or TorchScript for real-time defect detection in factories.

You might also like