The LAB color space has values from -128 to +128. Since neural networks are sensitive to the scale of input values, we normalize the transformed pixel values from -128 to +128 and bring them within the -1 to +1 range. The same is showcased in the following code snippet:

def prep_data(file_list=[],
              dir_path=None,
              dim_x=256,
              dim_y=256):

    #Get images

    X = []
for filename in file_list:
    X.append(img_to_array(
                        sp.misc.imresize(
                        load_img(
                        dir_path+filename),
                        (dim_x, dim_y))

           )

        )
    X = np.array(X, dtype=np.float64)
    X = 1.0/255*X
    return X

Once transformed, we then split the data into train and test sets. For splitting, we utilize train_test_split utility from sklearn.