# defect-detecting

**Repository Path**: wk_09/defect-detecting

## Basic Information

- **Project Name**: defect-detecting
- **Description**: Detect and classify defects in steels
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2020-03-28
- **Last Updated**: 2021-08-01

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Steel Defect Detection Project

This is a project that use U-Net to predict and classify the defect regions of steel images with Kaggle dataset [Severstal: Steel Defect Detection](https://www.kaggle.com/c/severstal-steel-defect-detection/overview/evaluation)  
  

## Exploring the data
The defect masks of the steel images are encoded using Run-length encoding. First we decoded the labels to masks and indicated the defect regions on the images.  
![1](https://github.com/RocioLiu/DefectDetection/blob/master/assets/0002cc93b.jpg)
![2](https://github.com/RocioLiu/DefectDetection/blob/master/assets/0007a71bf.jpg)
![3](https://github.com/RocioLiu/DefectDetection/blob/master/assets/000a4bcdd.jpg)
![4](https://github.com/RocioLiu/DefectDetection/blob/master/assets/000f6bf48.jpg)
![5](https://github.com/RocioLiu/DefectDetection/blob/master/assets/0014fce06.jpg)
  
  
## Training
We built a U-Net model and trained it for 30 epochs
![](https://github.com/RocioLiu/DefectDetection/blob/master/assets/unet.png)    
  
### Metrics
The metrics we use to evaluate our model is mean [Dice coefficient](https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient).

The training result:

![](https://github.com/RocioLiu/DefectDetection/blob/master/assets/training1.png)
   
   
## Prediction
We display the prediction of a batch of images  
   
![](https://github.com/RocioLiu/DefectDetection/blob/master/assets/prediction1.png)