self-directed research (2023 spring semester )

2023-1 Self-directed research 2023-1 자기주도연구1

I built a welding dataset from weld X-ray images. Using this dataset, we trained two models, the simple-CNN model and the resnet18 model, to classify the defective and normal areas in the weld X-ray images. To verify that the resulting models were properly trained, we used grad-CAM to examine the regions where classes are activated.

Requirements

Python 3.+

$ pip install opencv-python numpy pillow
$ pip install torch torchvision
$ pip install grad-cam

utility

getmean.py : get mean and std from welding dataset for normalization

training Simple CNN 2-layer model

train.py : training code of Simple CNN 2-layer model

test.py : testing code of Simple CNN 2-layer model

training Resnet18 model

train_resnet18.py : training code of Resnet18 model

test_resnet18.py : testing code of Resnet18 model

grad-CAM Visualization

test_cam_conv.py : applying gradCAM to Simple CNN 2-layer model for visualiztion

test_cam_resnet18.py : applying gradCAM to Resnet18 model for visualiztion