Deep Learning - HW5: Automated Optical Inspection (AOI)

Author: Jay Liao (re6094028@gs.ncku.edu.tw)

This is assignment 5 of Deep Learning, a course at Institute of Data Science, National Cheng Kung University. This project aims to utilize deep learning techniques to perform defect classification for AOI images.

Data

• Images: please go to the page of the competition on the AIdea here to download raw image files with two folders, ./train_images/ and ./test_images/.

• File name lists of images: ./train.csv and ./test.txt.

Code

• main_keras.py: the main program for training a 5-layered CNN with Keras. The following codes demonstrate the model structute.

raw_model = Sequential([
    layers.Conv2D(16, 3, input_shape=(size[0], size[1], 1), padding='same', activation='relu', kernel_regularizer=tf.keras.regularizers.l2(l=0.01)),
    layers.MaxPooling2D(),
    layers.Conv2D(24, 3, padding='same', activation='relu', kernel_regularizer=tf.keras.regularizers.l2(l=0.01)),
    layers.MaxPooling2D(),
    layers.Conv2D(24, 3, padding='same', activation='relu', kernel_regularizer=tf.keras.regularizers.l2(l=0.01)),
    layers.MaxPooling2D(),
    layers.Conv2D(24, 3, padding='same', activation='relu', kernel_regularizer=tf.keras.regularizers.l2(l=0.01)),
    layers.MaxPooling2D(),
    layers.Conv2D(24, 3, padding='same', activation='relu', kernel_regularizer=tf.keras.regularizers.l2(l=0.01)),
    layers.MaxPooling2D(),
    layers.Dropout(0.2),
    layers.Flatten(),
    layers.Dense(32, activation='relu'),
])

• Source codes for training a 5-layered CNN with Keras:

  • ./aoi_keras/args.py: define the arguments parser

  • ./aoi_keras/trainer.py: class for training, predicting, and evaluating the models

  • ./aoi_keras/load_data.py: functions for loading images with train/val/test splitting

• main_keras.py: the main program for fine tuning a pretrained ResNet50 model and an EfficientNet-b7 model.

• Source codes for fine tuning a pretrained ResNet50 model and an EfficientNet-b7 model:

  • ./aoi_torch/args.py: define the arguments parser

  • ./aoi_torch/trainer.py: class for training, predicting, and evaluating the models

  • ./aoi_torch/models.py: define the models.

  • ./aoi_torch/load_data.py: functions for loading images with train/val/test splitting

Folders

• ./train_images/ should contain 2,528 raw training image files (please go here to download). They will be splitted into three subsets after running the main program (folders ./images_tr/, ./images_va/, and ./images_te/ will be created in torch version).

• ./test_images/ should contain 10,142 testing image files without ground truth lables (please go here to download). The folder ./images_test/ will be created after running the main program of torch version.

• ./output_torch/ and ./output_keras/ will contain trained models, model performances, and experiments results after running.

Requirements

numpy
pandas
tqdm
opencv-python==3.4.2.16
matplotlib==3.1.3
torch
keras==2.2.5
tensorflow==2.3.1
tensorflow-gpu==2.3.1

Usage

1. Clone this repo.

git clone https://github.com/jayenliao/DL-AOI.git

2. Set up the required packages.

cd DL-AOI
pip3 install -H requirements.txt

3. Run the experiments.

python3 main_keras.py
python3 main_torch.py

Reference

1. https://github.com/bentrevett/pytorch-image-classification
2. https://github.com/pytorch/tutorials
3. https://github.com/pytorch/examples
4. https://colah.github.io/posts/2014-10-Visualizing-MNIST/
5. https://distill.pub/2016/misread-tsne/
6. https://towardsdatascience.com/visualising-high-dimensional-datasets-using-pca-and-t-sne-in-python-8ef87e7915b
7. https://github.com/activatedgeek/LeNet-5
8. https://github.com/ChawDoe/LeNet5-MNIST-PyTorch
9. https://github.com/kuangliu/pytorch-cifar
10. https://github.com/akamaster/pytorch_resnet_cifar10
11. https://sgugger.github.io/the-1cycle-policy.html