Classifying images

Image Modeling with Keras

Ariel Rokem

Senior Data Scientist, University of Washington

Image classification

Image classification: training

Image classification: evaluation

Representing class data: one-hot encoding

labels = ["shoe", "dress", "shoe", "t-shirt", 
          "shoe", "t-shirt", "shoe", "dress"]

Representing class data: one-hot encoding

array([[0., 0., 1.],    <= shoe
       [0., 1., 0.],    <= dress
       [0., 0., 1.],    <= shoe
       [1., 0., 0.],    <= t-shirt
       [0., 0., 1.],    <= shoe
       [1., 0., 0.],    <= t-shirt
       [0., 0., 1.],    <= shoe
       [0., 1., 0.]])   <= dress

One-hot encoding

categories = np.array(["t-shirt", "dress", "shoe"])

n_categories = 3
ohe_labels = np.zeros((len(labels), n_categories))

for ii in range(len(labels)):

    jj = np.where(categories == labels[ii])

    ohe_labels[ii, jj] = 1

One-hot encoding: testing predictions

test

array([[0., 0., 1.], 
       [0., 1., 0.], 
       [0., 0., 1.], 
       [0., 1., 0.], 
       [0., 0., 1.],
       [0., 0., 1.],
       [0., 0., 1.],
       [0., 1., 0.]])

(test * prediction).sum()

6.0

prediction

array([[0., 0., 1.], 
       [0., 1., 0.], 
       [0., 0., 1.], 
       [1., 0., 0.], <= incorrect
       [0., 0., 1.],
       [1., 0., 0.], <= incorrect
       [0., 0., 1.], 
       [0., 1., 0.]])

Let's practice!

Image Modeling with Keras