TensorFlow Hub モデルの OpenVINO 中間表現 (IR) への変換#

from pathlib import Path 
import os 
import requests 

os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2" 
os.environ["TF_USE_LEGACY_KERAS"] = "1" 
os.environ["TFHUB_CACHE_DIR"] = str(Path("./tfhub_modules").resolve()) 

import tensorflow_hub as hub 
import tensorflow as tf 
import PIL 
import numpy as np 
import matplotlib.pyplot as plt 

import openvino as ov 

tf.get_logger().setLevel("ERROR")

IMAGE_SHAPE = (224, 224) 
IMAGE_URL, IMAGE_PATH = ( 

"https://storage.googleapis.com/download.tensorflow.org/example_images/grace_hopper.jpg", "data/grace_hopper.jpg", 
) 
MODEL_URL, MODEL_PATH = ( 
    "https://www.kaggle.com/models/google/mobilenet-v1/frameworks/tensorFlow2/variations/100-224-classification/versions/2", 
    "models/mobilenet_v2_100_224.xml", 
)

model = hub.KerasLayer(MODEL_URL, input_shape=IMAGE_SHAPE + (3,))

Path(IMAGE_PATH).parent.mkdir(parents=True, exist_ok=True) 

r = requests.get(IMAGE_URL) 
with Path(IMAGE_PATH).open("wb") as f: 
    f.write(r.content) 
grace_hopper = PIL.Image.open(IMAGE_PATH).resize(IMAGE_SHAPE) 
grace_hopper

grace_hopper = np.array(grace_hopper) / 255.0 
grace_hopper.shape

if not Path(MODEL_PATH).exists(): 
    converted_model = ov.convert_model(model) 
    ov.save_model(converted_model, MODEL_PATH)

import ipywidgets as widgets 

core = ov.Core() 
device = widgets.Dropdown( 
    options=core.available_devices + ["AUTO"], 
    value="AUTO", 
    description="Device:", 
    disabled=False, 
) 

device

compiled_model = core.compile_model(MODEL_PATH, device_name=device.value)

output = compiled_model(grace_hopper[np.newaxis, ...])[0] 
output.shape

predicted_class = np.argmax(output[0], axis=-1) 
predicted_class

labels_path = tf.keras.utils.get_file( 
    "ImageNetLabels.txt", 
    "https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt", 
) 
imagenet_labels = np.array(open(labels_path).read().splitlines()) 
plt.imshow(grace_hopper) 
plt.axis("off") 
predicted_class_name = imagenet_labels[predicted_class] 
_ = plt.title("Prediction: " + predicted_class_name.title())

inputs = { 
    "placeholder": tf.keras.layers.Input(shape=(None, None, 3)), 
    "placeholder_1": tf.keras.layers.Input(shape=(None, None, 3)), 
} 
model = hub.KerasLayer(MODEL_URL, signature="serving_default", 
signature_outputs_as_dict=True) # 入力を辞書として渡すことを可能にする署名を定義 
outputs = model(inputs) 
model = tf.keras.Model(inputs=inputs, outputs=outputs)

if not Path(MODEL_PATH).exists():     Path(MODEL_PATH).parent.mkdir(parents=True, exist_ok=True) 
    converted_model = ov.convert_model(model) 
    ov.save_model(converted_model, MODEL_PATH)

import ipywidgets as widgets 

core = ov.Core() 
device = widgets.Dropdown( 
    options=core.available_devices + ["AUTO"], 
    value="AUTO", 
    description="Device:", 
    disabled=False, 
) 

device

compiled_model = core.compile_model(MODEL_PATH, device_name=device.value)

if not Path(STYLE_IMAGE_PATH).exists(): 
    r = requests.get(STYLE_IMAGE_URL) 
    with open(STYLE_IMAGE_PATH, "wb") as f: 
        f.write(r.content) 
if not Path(CONTENT_IMAGE_PATH).exists(): 
    r = requests.get(CONTENT_IMAGE_URL) 
    with open(CONTENT_IMAGE_PATH, "wb") as f: 
        f.write(r.content) 

def load_image(dst): 
    image = cv2.imread(dst) 
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) # 画像の色を RGB 空間に変換 
    image = image / 255 # [0, 1] 区間に正規化 
    image = image.astype(np.float32) 
    return image

content_image = load_image(CONTENT_IMAGE_PATH) 
style_image = load_image(STYLE_IMAGE_PATH) 
style_image = cv2.resize(style_image, (256, 256)) # モデルは 256x256 の画像でトレーニング

result = compiled_model([content_image[np.newaxis, ...], style_image[np.newaxis, ...]])[0]

title2img = { 
    "Source image": content_image, 
    "Reference style": style_image, 
    "Result": result[0], 
} 
plt.figure(figsize=(12, 12)) 
for i, (title, img) in enumerate(title2img.items()): 
    ax = plt.subplot(1, 3, i + 1) 
    ax.set_title(title) 
    plt.imshow(img) 
    plt.axis("off")