CNN_Auto_Ownership_Prediction

By Shen Hao Wang and Jason Lu

Paper is under jasonllu.pdf

Preprocessing notebooks:

Gaussian Pyramid Preprocess.ipynb data_preprocessing_notebook_oldImgs.ipynb

Final Models:

nn_GPU_8fc.ipynb - Feed Forward Network with NHTS dataset 9 layers cnn_GPU_ADAM_2Conv.ipynb - CNN with satellite images CNN_Gaussian_Pyramid_2conv.ipynb - CNN with Gaussian Pyramid (shared weights) CNN_Gaussian_Pyramid_2conv_unshared.ipynb - CNN with Gaussian Pyramid (unshared weights) aggregatedNNandCNN_GPU_2conv.ipynb - CNN with satellite images augmented with NHTS dataset aggregatedCNN_Gaussian_Pyramid_2conv.ipynb - CNN with Gaussian Pyramid augmented with NHTS dataset (shared weights) aggregatedNNandCNN_GPU_2conv_unshared.ipynb - CNN with Gaussian Pyramid augmented with NHTS dataset (unshared weights)

Filter Visualizations (Representation Learning):

filter_visualization_cnn_net.ipynb

Other models that we tried that resulted in lower accuracies (more layers):

aggregatedCNN_Gaussian_Pyramid_6layer.ipynb aggregatedNNandCNN_GPU_6layer.ipynb nn_GPU_3fc.ipynb