/Segmentation-based-deep-learning-approach-for-surface-defect-detection

Segmentation-based deep-learning approach for surface-defect detection with pytorch

Primary LanguagePythonMIT LicenseMIT

Segmentation-Based Deep-Learning Approach for Surface-Defect Detection

Try to make this as a productive project(on going)

Implement this paper by Pytorch

SDASDD

Network Arch

network arch

usage

prepare your KolektorSDD dataset

1. under KolktorSDD dir, create two txt files('train.txt','val.txt')
2. write the image filename and label filename like below:
```
kos01/Part5.jpg kos01/Part5_label.bmp
kos02/Part6.jpg kos02/Part6_label.bmp
kos03/Part2.jpg kos03/Part2_label.bmp
kos04/Part3.jpg kos04/Part3_label.bmp
...
```

start training

1. modify 'train.py'
DATAROOT
GLOBALEPOCH
INPUTHW
2. python train.py

pre-trained model(google drive)

  1. pre-trained segment net
  2. pre-trained decision net

TODO

  • Forward finished
  • Segmentation Net Training & Validate functions
  • Decision Net Training & Validate functions
  • resume segmentation net training script
  • resume decision net training script
  • Tensorboard record
  • Model fuse
  • ONNX format
  • Windows deployment
  • Linux deployment