|
import torch |
|
import torch.nn as nn |
|
import torchvision.models as models |
|
import torchvision.transforms as transforms |
|
from PIL import Image |
|
import numpy as np |
|
import gradio as gr |
|
import tensorflow as tf |
|
import tensorflow_hub as hub |
|
import cv2 |
|
|
|
|
|
IMAGE_SIZE = (256, 256) |
|
style_transfer_model = hub.load('https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2') |
|
|
|
|
|
vgg = models.vgg19(weights=models.VGG19_Weights.IMAGENET1K_V1).features |
|
vgg.eval() |
|
|
|
|
|
def load_image_pytorch(image, transform=None): |
|
if transform: |
|
image = transform(image).unsqueeze(0) |
|
return image |
|
|
|
|
|
transform = transforms.Compose([ |
|
transforms.Resize((224, 224)), |
|
transforms.ToTensor(), |
|
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]), |
|
]) |
|
|
|
|
|
def deep_dream(model, image, iterations, lr): |
|
for i in range(iterations): |
|
image.requires_grad = True |
|
features = model(image) |
|
loss = features.norm() |
|
loss.backward() |
|
with torch.no_grad(): |
|
image += lr * image.grad |
|
image = image.clamp(0, 1) |
|
image.grad = None |
|
return image |
|
|
|
|
|
def load_image_tensorflow(image): |
|
image = cv2.resize(image, IMAGE_SIZE, interpolation=cv2.INTER_AREA) |
|
image = image.astype(np.float32)[np.newaxis, ...] / 255. |
|
if image.shape[-1] == 4: |
|
image = image[..., :3] |
|
return image |
|
|
|
|
|
def apply_sharpness(image, intensity): |
|
kernel = np.array([[0, -intensity, 0], |
|
[-intensity, 1 + 4 * intensity, -intensity], |
|
[0, -intensity, 0]]) |
|
sharp_image = cv2.filter2D(image, -1, kernel) |
|
return np.clip(sharp_image, 0, 255) |
|
|
|
|
|
def interpolate_images(baseline, target, alpha): |
|
return baseline + alpha * (target - baseline) |
|
|
|
|
|
def combine_images(image1, image2, weight1, style_density, content_sharpness): |
|
|
|
image1_pil = Image.fromarray(image1) |
|
image2_pil = Image.fromarray(image2) |
|
|
|
image1_pytorch = load_image_pytorch(image1_pil, transform) |
|
image2_pytorch = load_image_pytorch(image2_pil, transform) |
|
|
|
|
|
weight2 = 1 - weight1 |
|
mixed_image_pytorch = weight1 * image1_pytorch + weight2 * image2_pytorch |
|
|
|
|
|
mixed_image_pytorch = deep_dream(vgg, mixed_image_pytorch, iterations=20, lr=0.01) |
|
|
|
|
|
mixed_image_pytorch = mixed_image_pytorch.squeeze(0).permute(1, 2, 0).detach().numpy() |
|
mixed_image_pytorch = (mixed_image_pytorch * 255).astype(np.uint8) |
|
|
|
|
|
content_image = load_image_tensorflow(mixed_image_pytorch) |
|
style_image = load_image_tensorflow(image2) |
|
|
|
|
|
content_image_sharp = apply_sharpness(content_image[0], intensity=content_sharpness) |
|
content_image_sharp = content_image_sharp[np.newaxis, ...] |
|
|
|
|
|
stylized_image = style_transfer_model(tf.constant(content_image_sharp), tf.constant(style_image))[0] |
|
|
|
|
|
stylized_image = interpolate_images(baseline=content_image[0], target=stylized_image.numpy(), alpha=style_density) |
|
stylized_image = np.array(stylized_image * 255, np.uint8) |
|
stylized_image = np.squeeze(stylized_image) |
|
|
|
return stylized_image |
|
|
|
|
|
example1 = np.array(Image.open("example1.jpg")) |
|
example2 = np.array(Image.open("example2.jpg")) |
|
|
|
|
|
interface = gr.Interface( |
|
fn=combine_images, |
|
inputs=[ |
|
gr.Image(type="numpy", label="Imagem 1"), |
|
gr.Image(type="numpy", label="Imagem 2"), |
|
gr.Slider(minimum=0, maximum=1, value=0.5, label="Peso da Imagem 1"), |
|
gr.Slider(minimum=0, maximum=1, value=0.5, label="Densidade do Estilo"), |
|
gr.Slider(minimum=0, maximum=10, value=1, label="Nitidez do Conteúdo") |
|
], |
|
outputs="image", |
|
title="Combinação de Imagens com Deep Dreaming e Transferência de Estilo", |
|
description="Ajuste os pesos e a densidade do estilo para combinar e estilizar as imagens. *v1.0 experimental 07/07/2024", |
|
examples=[ |
|
["example1.jpg", "example2.jpg", 0.5, 0.5, 1] |
|
] |
|
) |
|
|
|
interface.launch() |
|
|
