StrikingLoo

fc_layer_size = 256
img_size = IMG_SIZE

conv_inputs = keras.Input(shape=(img_size[1], img_size[0],3), name='ani_image')
conv_layer = layers.Conv2D(128, kernel_size=3, activation='relu')(conv_inputs)
conv_layer = layers.MaxPool2D(pool_size=(2,2))(conv_layer)

conv_layer = layers.Conv2D(128, kernel_size=3, activation='relu')(conv_layer)
conv_layer = layers.MaxPool2D(pool_size=(2,2))(conv_layer)

StrikingLoo

cat_quantity = sum(labels_valid)

for i in range(1,10):
    print('threshold :'+str(.1*i))
    print(sum(labels_valid[preds > .1*i])/labels_valid[preds > .1*i].shape[0])

StrikingLoo

fc_layer_size = 256
img_size = IMG_SIZE

conv_inputs = keras.Input(shape=(img_size[1], img_size[0],3), name='ani_image')
#first convolutional layer.
conv_layer = layers.Conv2D(48, kernel_size=3, activation='relu')(conv_inputs)
conv_layer = layers.MaxPool2D(pool_size=(2,2))(conv_layer)
#second convolutional layer.
conv_layer = layers.Conv2D(48, kernel_size=3, activation='relu')(conv_layer)
conv_layer = layers.MaxPool2D(pool_size=(2,2))(conv_layer)

StrikingLoo

preds = conv_model.predict(x_valid)
preds = np.asarray([pred[0] for pred in preds])
np.corrcoef(preds, labels_valid)[0][1] # 0.15292172

StrikingLoo

fc_layer_size = 128
img_size = IMG_SIZE

conv_inputs = keras.Input(shape=(img_size[1], img_size[0],3), name='ani_image')
conv_layer = layers.Conv2D(24, kernel_size=3, activation='relu')(conv_inputs)
conv_layer = layers.MaxPool2D(pool_size=(2,2))(conv_layer)
conv_x = layers.Flatten(name = 'flattened_features')(conv_layer) #turn image to vector.

conv_x = layers.Dense(fc_layer_size, activation='relu', name='first_layer')(conv_x)
conv_x = layers.Dense(fc_layer_size, activation='relu', name='second_layer')(conv_x)

StrikingLoo

customAdam = keras.optimizers.Adam(lr=0.001)
model.compile(optimizer=customAdam,  # Optimizer
              # Loss function to minimize
              loss="mean_squared_error",
              # List of metrics to monitor
              metrics=["binary_crossentropy","mean_squared_error"])
              
print('# Fit model on training data')

history = model.fit(x_train, 

StrikingLoo

from tensorflow import keras
from tensorflow.keras import layers

total_pixels = img_size[0] *img_size[1] * 3
fc_size = 512

inputs = keras.Input(shape=(img_size[1], img_size[0],3), name='ani_image')
x = layers.Flatten(name = 'flattened_img')(inputs) #turn image to vector.

x = layers.Dense(fc_size, activation='relu', name='first_layer')(x)

StrikingLoo

SAMPLE_SIZE = 2048
print("loading training cat images...")
cat_train_set = np.asarray([pixels_from_path(cat) for cat in glob.glob('cats/*')[:SAMPLE_SIZE]])
print("loading training dog images...")
dog_train_set = np.asarray([pixels_from_path(dog) for dog in glob.glob('dogs/*')[:SAMPLE_SIZE]])

valid_size = 512
print("loading validation cat images...")
cat_valid_set = np.asarray([pixels_from_path(cat) for cat in glob.glob('cats/*')[-valid_size:]])
print("loading validation dog images...")

StrikingLoo

IMG_SIZE = (94, 125)
def pixels_from_path(file_path):
    im = Image.open(file_path)
    im = im.resize(IMG_SIZE)
    np_im = np.array(im)
    #matrix of pixel RGB values
    return np_im

StrikingLoo

shape_counts = defaultdict(int)
for i, cat in enumerate(glob.glob('cats/*')[:1000]):
    if i%100==0:
        print(i)
    img_shape = pixels_from_path(cat).shape #loads image as np matrix and checks shape.
    shape_counts[str(img_shape)]= shape_counts[str(img_shape)]+ 1
    
shape_items = list(shape_counts.items())
shape_items.sort(key = lambda x: x[1])
shape_items.reverse()