doleron

# adapted from: https://pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/
def intersection_over_union(boxA, boxB):
	xA = max(boxA[0], boxB[0])
	yA = max(boxA[1], boxB[1])
	xB = min(boxA[0] + boxA[2], boxB[0] + boxB[2])
	yB = min(boxA[1] + boxA[3], boxB[1] + boxB[3])
	interArea = max(0, xB - xA + 1) * max(0, yB - yA + 1)
	boxAArea = (boxA[2] + 1) * (boxA[3] + 1)
	boxBArea = (boxB[2] + 1) * (boxB[3] + 1)
	iou = interArea / float(boxAArea + boxBArea - interArea)

doleron

plt.figure(figsize=(20, 10))
for images, labels in train_ds.take(1):
    for i in range(BATCH_SIZE):
        ax = plt.subplot(4, BATCH_SIZE//4, i + 1)
        label = labels[0][i]
        box = (labels[1][i] * input_size)
        box = tf.cast(box, tf.int32)

        image = images[i].numpy().astype("float") * 255.0
        image = image.astype(np.uint8)

doleron

input_size = 244

def format_image(img, box):
    height, width = img.shape 
    max_size = max(height, width)
    r = max_size / input_size
    new_width = int(width / r)
    new_height = int(height / r)
    new_size = (new_width, new_height)
    resized = cv.resize(img, new_size, interpolation= cv.INTER_LINEAR)

doleron

import os, random

def list_files(full_data_path = "data/obj/", image_ext = '.jpg', split_percentage = [70, 20]):
    
    files = []

    discarded = 0
    masked_instance = 0

    for r, d, f in os.walk(full_data_path):

doleron

model = build_model(tf.keras.layers.Input(shape=(input_size, input_size, 1,)))

model.compile(optimizer=tf.keras.optimizers.Adam(), 
    loss = {'classifier_head' : 'categorical_crossentropy', 'regressor_head' : 'mse' }, 
    metrics = {'classifier_head' : 'accuracy', 'regressor_head' : 'mse' })
    
EPOCHS = 100
BATCH_SIZE = 32

doleron

def build_feature_extractor(inputs):

    x = tf.keras.layers.Conv2D(16, kernel_size=3, activation='relu', input_shape=(input_size, input_size, 1))(inputs)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

    x = tf.keras.layers.Conv2D(32, kernel_size=3, activation = 'relu')(x)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

    x = tf.keras.layers.Conv2D(64, kernel_size=3, activation = 'relu')(x)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

doleron

def build_regressor(inputs):

    x = tf.keras.layers.Conv2D(16, kernel_size=3, activation='relu', input_shape=(input_size, input_size, 1))(inputs)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

    x = tf.keras.layers.Conv2D(32, kernel_size=3, activation = 'relu')(x)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

    x = tf.keras.layers.Conv2D(64, kernel_size=3, activation = 'relu')(x)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

doleron

CLASSES = 2

def build_classifier(inputs):

    x = tf.keras.layers.Conv2D(16, kernel_size=3, activation='relu', input_shape=(input_size, input_size, 1))(inputs)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

    x = tf.keras.layers.Conv2D(32, kernel_size=3, activation = 'relu')(x)
    x = tf.keras.layers.AveragePooling2D(2,2)(x)

doleron

CLASSES = 2
model = tf.keras.models.Sequential([
    
    tf.keras.layers.Conv2D(16, kernel_size=3, activation='relu', input_shape=(224, 224, 1)),
    tf.keras.layers.AveragePooling2D(2,2),
    
    tf.keras.layers.Conv2D(32, kernel_size=3, activation = 'relu'),
    tf.keras.layers.AveragePooling2D(2,2),
    
    tf.keras.layers.Conv2D(64, kernel_size=3, activation = 'relu'),

doleron

function mse(model, data) {
    let result = 0.0;
    for (let i = 0; i < data.length; ++i) {
      	const elem = data[i];
      	const x = elem.x;
        const y = elem.y;
        const y_hat = predict(x, model);
        const diff = y_hat - y;
        result += diff * diff;
    }