Create/load datasets:

Download a bunch of different datasets:

• ADFES, KDEF, MMI, CK+ (all available online with educational license), and FER if you want to do CNN training

For CNN dataset:

Many scripts to create 48x48 images out of the given datasets.

For nonCNN dataset:

Run nonCNN_unbalanced_dataset.m to get a smaller but higher-resolution, more balanced dataset Related scripts before running nonCNN_unbalanced_dataset: load_ADFES_take_away_disgust.m and load_MMI_take_away_disgust.m (requires download in certain folders of different datasets)

non-CNN methods

The non-CNN methods are executed in MATLAB.

MATLAB toolboxes to download

heatmap toolbox for confusion matrices

Tests to Run

For training:

Run trainClassifier.m to get back a trained classifier and display a confusion matrix for you! - options to change from LDA to SVM.

run gaussian_blur.m to generate topX.mat and bottomX.mat (blurred images)

run model_top_bottom_testing.m to now test your model given blurred images.

change trainClassifier.m if you want to also train a model with blurred images, and then test with blurred images.

Naming Conventions

MDL_*s*o*.mat = type, sizes, orientations

• the model classifier MDL_comps_*s*o*.mat = type, sizes, orientations
• the components for this certain model (for transformation purposes)

FOR CAFFE TRAINING AUGMENTED LAYERS:

make sure you put this in your bash: export PYTHONPATH=~/path/to/cs_99_code/PYTHON/InvertedLayerRandom:$PYTHONPATH so that when you run caffe train, you can add the inverted layer in!

Deprecated Files

loadImages.m