Quick, Draw! dataset from google, with added noise, classification amongst 30 categories or "empty" image.
For each iPython Notebook, it is sufficient to run each cell sequentially in order to replicate our results, as detailed in our report on Gradescope. Each notebook is independent. The data is available if contacted.
all/test_images.npy Test images
all/train_images.npy Train images
all/train_labels.csv Train labels
Kaggle_Project_CNN.ipynb Best performing CNN model contained in this notebook
SVM_Grid_Search.ipynb Contains SVM on PCA, with a grid search for hyper parameters
Backprop.ipynb Contains hand-coded neural network back-propogation class
SIFT_SVM.ipynb SIFT features in SVM
SVM.ipynb Simple SVM on the data
Keras OpenCV 3+ sklearn pandas numpy matplotlib multiprocessing datetime io os