Anomally_Detection_Partially_Finetuning_VGG19_PyTorch

Perform anomally detection on a dataset containing wall crack by partially finetuning a pre-trained VGG19 model in PyTorch Partially fine-tuned a pre-trained VGG19 model on a data-augmented training dataset containing images of walls containing cracks and no cracks to obtain a test accuracy of 92% for detecting crack in the test images and localized the cracks afterwards using activation maps from the last convolution layer of the VGG19 model and by projecting the activation maps after reshaping it by linear-up sampling and taking dot product with the corresponding class weight of the last fully connected layer of the modified 'classifier' module of the VGG19 model to figure out the activation for our predicted class.