Steel-Defect-Detection

This is an end to end Image Segmentation case study for Steel Defect Detection

Dataset: Severstal: Steel Defect Detection | kaggle

You can aslo refer my blog on 'Medium' about this case-study work. Blog Link: Steel Defect Detection — Image Segmentation using TensorFlow

Contents:

EDA & Data Preparation.ipynb

1. Business Problem

  1.1 Description
  1.2 Sources/Useful Links
  1.3 Real world/Business Objectives and Constraints

2. Machine Learning Probelm

  2.1 Data
    2.1.1 Data Overview
  2.2 Mapping the real world problem to an ML problem
    2.2.1 Type of Machine Leaning Problem
    2.2.2 Performance Metric

3. Exploratory data analysiss

4. Data Preperation

UNET.ipynb

5. UNET

  5.1 Loading Data
  5.2 Data Generator Implementation
  5.3 Utility Functions
  5.4 Defining Unet Architecture
  5.5 Traing Model
  5.6 Visualizing Model Predictions
  5.7 Preparing Data for submission
  5.8 Kaggle Submission Score

Linknet.ipynb

6. LINKNET

  6.1 Loading Data
  6.2 Data Generator Implementation
  6.3 Utility Functions
  6.4 Defining LinkNet Architecture
  6.5 Traing Model
  6.6 Visualizing Model Predictions
  6.7 Preparing Data for submission
  6.8 Kaggle Submission Score

UNet_with_ResNet.ipynb

7. UNet with Resnet

  7.1 Loading Data
  7.2 Data Generator Implementation
  7.3 Utility Functions
  7.4 Defining Unet Architecture with ResNet as backbone
  7.5 Traing Model
  7.6 Visualizing Model Predictions
  7.7 Preparing Data for submission
  7.8 Kaggle Submission Score

LinkNet_with_ResNet.ipynb

8. LinkNet with Resnet

  8.1 Loading Data
  8.2 Data Generator Implementation
  8.3 Utility Functions
  8.4 Defining Linknet Architecture with ResNet as backbone
  8.5 Traing Model(1-30 epochs)
  8.6 Traing Model(31-60 epochs)
  8.7 Visualizing Model Predictions
  8.8 Preparing Data for submission
  8.9 Kaggle Submission Score

UNET_plusplus.ipynb

9. UNET++

  9.1 Loading Data
  9.2 Data Generator Implementation
  9.3 Utility Functions
  9.4 Defining Unet++ Architecture
  9.5 Traing Model(1-30 epochs)
  9.6 Traing Model(31-60 epochs)
  9.7 Visualizing Model Predictions
  9.8 Preparing Data for submission
  9.9 Kaggle Submission Score

Results.ipynb

10. Results

Final.ipynb

11. Final Functions

  11.1 Final Function - 1
  11.2 Final Function - 2

Quantization.ipynb

12. Post-Training Quantization

  12.1 Quantization
  12.2 Size Comparision
  12.3 performance comparision