import math, requests from pathlib import Path import tensorflow as tf from tensorflow.keras import Model from tensorflow.keras.models import Sequential from tensorflow.keras.layers import ( Dense, InputLayer, Dropout, Conv1D, Flatten, Reshape, MaxPooling1D, BatchNormalization, Conv2D, GlobalMaxPooling2D, Lambda, GlobalAveragePooling2D) from tensorflow.keras.optimizers.legacy import Adam, Adadelta from tensorflow.keras.losses import categorical_crossentropy from ei_tensorflow.velo import train_keras_model_with_velo sys.path.append('./resources/libraries') import ei_tensorflow.training WEIGHTS_PATH = './transfer-learning-weights/keras/mobilenet_2_5_128_tf.h5' # Download the model weights root_url = 'https://cdn.edgeimpulse.com/' p = Path(WEIGHTS_PATH) if not p.exists(): print(f"Pretrained weights {WEIGHTS_PATH} unavailable; downloading...") if not p.parent.exists(): p.parent.mkdir(parents=True) weights_data = requests.get(root_url + WEIGHTS_PATH[2:]).content with open(WEIGHTS_PATH, 'wb') as f: f.write(weights_data) print(f"Pretrained weights {WEIGHTS_PATH} unavailable; downloading OK") print("") INPUT_SHAPE = (96, 96, 3) base_model = tf.keras.applications.MobileNet( input_shape = INPUT_SHAPE, weights = WEIGHTS_PATH, alpha = 0.25 ) base_model.trainable = False model = Sequential() model.add(InputLayer(input_shape=INPUT_SHAPE, name='x_input')) # Don't include the base model's top layers last_layer_index = -5 model.add(Model(inputs=base_model.inputs, outputs=base_model.layers[last_layer_index].output)) model.add(Reshape((-1, model.layers[-1].output.shape[3]))) model.add(Dropout(0.1)) model.add(Flatten()) model.add(Dense(classes, activation='softmax')) # Implements the data augmentation policy def augment_image(image, label): # Flips the image randomly image = tf.image.random_flip_left_right(image) # Increase the image size, then randomly crop it down to # the original dimensions resize_factor = random.uniform(1, 1.2) new_height = math.floor(resize_factor * INPUT_SHAPE[0]) new_width = math.floor(resize_factor * INPUT_SHAPE[1]) image = tf.image.resize_with_crop_or_pad(image, new_height, new_width) image = tf.image.random_crop(image, size=INPUT_SHAPE) # Vary the brightness of the image image = tf.image.random_brightness(image, max_delta=0.2) return image, label train_dataset = train_dataset.map(augment_image, num_parallel_calls=tf.data.AUTOTUNE) BATCH_SIZE = args.batch_size or 256 EPOCHS = args.epochs or 20 # If True, non-deterministic functions (e.g. shuffling batches) are not used. # This is False by default. ENSURE_DETERMINISM = args.ensure_determinism if not ENSURE_DETERMINISM: train_dataset = train_dataset.shuffle(buffer_size=BATCH_SIZE*4) prefetch_policy = 1 if ENSURE_DETERMINISM else tf.data.AUTOTUNE train_dataset = train_dataset.batch(BATCH_SIZE, drop_remainder=False).prefetch(prefetch_policy) validation_dataset = validation_dataset.batch(BATCH_SIZE, drop_remainder=False).prefetch(prefetch_policy) callbacks.append(BatchLoggerCallback(BATCH_SIZE, train_sample_count, epochs=EPOCHS, ensure_determinism=ENSURE_DETERMINISM)) train_keras_model_with_velo( keras_model=model, training_data=train_dataset, validation_data=validation_dataset, loss_fn=tf.keras.metrics.categorical_crossentropy, num_epochs=EPOCHS, callbacks=callbacks )