miwaliu/SbeNet-for-Defect-Detection

The office implementation of Two-stage Deep Neural Network with Joint Loss and Multi-level Representations for Defect Detection

SbeNet-for-Defect-Detection

1. Introduction

This repository is the offical implementation of Two-stage Deep Neural Network with Joint Loss and Multi-level Representations for Defect Detection. Up to now, we have uploaded the code and the pre-trained model.

2. Note

1. To reproduce the results in our paper, please construct the Data folder indicated in DataStructure.txt.
2. For compatibility reasons, the code is recommended to be run under Linux environment.

3. Usage

1. Modify init.ini according to your device conditions.
2. Download the dataset and construct the Data folder indicated in DataStructure.txt. For convenience we provide QuickSplit.py to build Data folder quickly (refer to Section 4).
3. We have provided quick-star scripts in folder RunningScript . Please run the script as follows:

sh RunningScript/Test_KSDD_F0.sh

3. We use tensorboard to record the results, please use the following command to view it:

tensorboard --logdir 'Model/XXXX/Log'

Where XXXX is the results storage folder, like KSDD_F0_lambda0.7_Test.

4. Build Data Folder

1. Down load and unpack the dataset. Please ensure the structure of dataset folder as follow:

   KSDD/
    ├── kos01
    ├── kos02
    ├── kos03
    ├──   .
    ├──   .
    ├──   .
    └── kos50
   ----------------------------
   Other/DAGM/
    ├── Class1
    │   ├── Test
    │   │   └── Label
    │   └── Train
    │       └── Label
    ├── Class2
    │   ├── Test
    │   │   └── Label
    │   └── Train
    │       └── Label
    │    .
    │    .
    │    .
    └── Class6
        ├── Test
        │   └── Label
        └── Train
            └── Label
    ----------------------------
    SSD/
    ├── sample_submission.csv
    ├── test_images
    ├── train.csv
    └── train_images
    
   

2. Run the QuickSplit.py as follos:

   python QuickSplit.py --Dataset 'KSDD' --Fold 0 --Dataset_dir 'KSDD' --Splitfile_dir 'Data_Split/KSDD_F0_Split.pkl'
   

   Where the split file for each datasets are provided in Data_Split .