README.md

metadata

license: cc
tags:
  - tensorflow
  - image-to-image

Whitebox Cartoonizer

Whitebox Cartoonizer [1] model in the SavedModel format. The model was exported to the SavedModel format using this notebook. Original model repository can be found here.

Inference code

import cv2
import numpy as np
import requests
import tensorflow as tf
from huggingface_hub import snapshot_download
from PIL import Image


def resize_crop(image):
    h, w, c = np.shape(image)
    if min(h, w) > 720:
        if h > w:
            h, w = int(720 * h / w), 720
        else:
            h, w = 720, int(720 * w / h)
    image = cv2.resize(image, (w, h), interpolation=cv2.INTER_AREA)
    h, w = (h // 8) * 8, (w // 8) * 8
    image = image[:h, :w, :]
    return image


def download_image(url):
    image = Image.open(requests.get(url, stream=True).raw)
    image = image.convert("RGB")
    image = np.array(image)
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    return image


def preprocess_image(image):
    image = resize_crop(image)
    image = image.astype(np.float32) / 127.5 - 1
    image = np.expand_dims(image, axis=0)
    image = tf.constant(image)
    return image


# Load the model and extract concrete function.
model_path = snapshot_download("sayakpaul/whitebox-cartoonizer")
loaded_model = tf.saved_model.load(model_path)
concrete_func = loaded_model.signatures["serving_default"]

# Download and preprocess image.
image_url = "https://huggingface.co./spaces/sayakpaul/cartoonizer-demo-onnx/resolve/main/mountain.jpeg"
image = download_image(image_url)
preprocessed_image = preprocess_image(image)

# Run inference.
result = concrete_func(preprocessed_image)["final_output:0"]

# Post-process the result and serialize it.
output = (result[0].numpy() + 1.0) * 127.5
output = np.clip(output, 0, 255).astype(np.uint8)
output = cv2.cvtColor(output, cv2.COLOR_BGR2RGB)
output_image = Image.fromarray(output)
output_image.save("result.png")

References

[1] Learning to Cartoonize Using White-box Cartoon Representations; Xinrui Wang and Jinze Yu; CVPR 2020.