Descrição das mudanças

app.py

@@ -2,6 +2,7 @@ import tensorflow as tf
 import tensorflow_hub as hub
 import numpy as np
 import gradio as gr
+import cv2
 
 # Carrega o modelo de transferência de estilo pré-treinado
 style_transfer_model = hub.load('https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2')
@@ -13,13 +14,34 @@ def load_image(image):
         image = image[..., :3]
     return image
 
-def style_transfer(content_image, style_image):
+def interpolate_images(baseline, target, alpha):
+    return baseline + alpha * (target - baseline)
+
+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 style_transfer(content_image, style_image, style_density, content_sharpness):
     # Processa as imagens
     content_image = load_image(content_image)
     style_image = load_image(style_image)
 
+    # Aplica nitidez na imagem de conteúdo antes da transferência de estilo
+    content_image_sharp = apply_sharpness(content_image[0], intensity=content_sharpness)
+    content_image_sharp = content_image_sharp[np.newaxis, ...]
+
     # Executa a transferência de estilo
-    stylized_image = style_transfer_model(tf.constant(content_image), tf.constant(style_image))[0]
+    stylized_image = style_transfer_model(tf.constant(content_image_sharp), tf.constant(style_image))[0]
+
+    # Interpola entre a imagem de conteúdo e a imagem estilizada para densidade de estilo
+    stylized_image = interpolate_images(
+        baseline=content_image[0],
+        target=stylized_image.numpy(),
+        alpha=style_density
+    )
 
     # Converte a imagem resultante para o formato correto
     stylized_image = np.array(stylized_image * 255, np.uint8)
@@ -30,8 +52,13 @@ def style_transfer(content_image, style_image):
 
 iface = gr.Interface(
     fn=style_transfer,
-    inputs=["image", "image"],
-    outputs="image"
+    inputs=[
+        gr.Image(label="Content Image"),  # Imagem de conteúdo
+        gr.Image(label="Style Image"),    # Imagem de estilo
+        gr.Slider(minimum=0, maximum=1, value=0.5, label="Adjust Style Density"),
+        gr.Slider(minimum=0, maximum=1, value=0.5, label="Content Sharpness")
+    ],
+    outputs=gr.Image(label="Stylized Image")
 )
 
-iface.launch()
+iface.launch()

requirements.txt

@@ -2,5 +2,6 @@ gradio
 tensorflow-hub
 numpy
 tensorflow
+opencv-python-headless