ok
Browse files- MEN-Denim-id_00000089-17_4_full.png +0 -0
- ben.jpg +0 -0
- image.webp +0 -0
- mode_app.py +232 -0
MEN-Denim-id_00000089-17_4_full.png
ADDED
|
ben.jpg
ADDED
|
image.webp
ADDED
|
mode_app.py
ADDED
|
@@ -0,0 +1,232 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import numpy as np
|
| 2 |
+
import gradio as gr
|
| 3 |
+
import cv2
|
| 4 |
+
import colorsys
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
def generate_distinct_colors(k):
|
| 8 |
+
colors = []
|
| 9 |
+
for i in range(k):
|
| 10 |
+
hue = i / k
|
| 11 |
+
saturation = 0.8
|
| 12 |
+
value = 0.8
|
| 13 |
+
rgb = colorsys.hsv_to_rgb(hue, saturation, value)
|
| 14 |
+
colors.append(np.array(rgb) * 255)
|
| 15 |
+
return np.array(colors, dtype=np.uint8)
|
| 16 |
+
|
| 17 |
+
def k_means_segmentation(image, k, max_iters=100, tol=1e-4):
|
| 18 |
+
pixels = image.reshape(-1, 3).astype(np.float32)
|
| 19 |
+
np.random.seed(42)
|
| 20 |
+
centroids = pixels[np.random.choice(pixels.shape[0], k, replace=False)]
|
| 21 |
+
|
| 22 |
+
for iteration in range(max_iters):
|
| 23 |
+
distances = np.linalg.norm(pixels[:, np.newaxis] - centroids, axis=2)
|
| 24 |
+
labels = np.argmin(distances, axis=1)
|
| 25 |
+
|
| 26 |
+
new_centroids = np.array([
|
| 27 |
+
pixels[labels == i].mean(axis=0) if np.sum(labels == i) > 0
|
| 28 |
+
else centroids[i]
|
| 29 |
+
for i in range(k)
|
| 30 |
+
])
|
| 31 |
+
|
| 32 |
+
if np.linalg.norm(new_centroids - centroids) < tol:
|
| 33 |
+
print(f"K-Means convergé après {iteration + 1} itérations.")
|
| 34 |
+
break
|
| 35 |
+
|
| 36 |
+
centroids = new_centroids
|
| 37 |
+
distinct_colors = generate_distinct_colors(k)
|
| 38 |
+
segmented_pixels = distinct_colors[labels].astype(np.uint8)
|
| 39 |
+
segmented_image = segmented_pixels.reshape(image.shape)
|
| 40 |
+
|
| 41 |
+
return segmented_image
|
| 42 |
+
|
| 43 |
+
def mean_shift_segmentation(image, bandwidth=30, max_iters=20, tol=1e-3, max_image_size=200):
|
| 44 |
+
"""
|
| 45 |
+
Version optimisée de la segmentation Mean Shift
|
| 46 |
+
|
| 47 |
+
Args:
|
| 48 |
+
image: Image d'entrée BGR
|
| 49 |
+
bandwidth: Rayon de recherche
|
| 50 |
+
max_iters: Nombre maximum d'itérations
|
| 51 |
+
tol: Seuil de convergence
|
| 52 |
+
max_image_size: Taille maximale de l'image (le plus grand côté)
|
| 53 |
+
"""
|
| 54 |
+
# Redimensionnement de l'image si nécessaire
|
| 55 |
+
h, w = image.shape[:2]
|
| 56 |
+
scale = max_image_size / max(h, w)
|
| 57 |
+
if scale < 1:
|
| 58 |
+
new_size = (int(w * scale), int(h * scale))
|
| 59 |
+
image = cv2.resize(image, new_size)
|
| 60 |
+
|
| 61 |
+
# Conversion en LAB
|
| 62 |
+
lab_image = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)
|
| 63 |
+
|
| 64 |
+
# Préparation des données
|
| 65 |
+
pixels = lab_image.reshape(-1, 3).astype(np.float32)
|
| 66 |
+
|
| 67 |
+
# Sous-échantillonnage pour les centres initiaux (1 pixel sur 10)
|
| 68 |
+
step = 10
|
| 69 |
+
centers = pixels[::step].copy()
|
| 70 |
+
|
| 71 |
+
# Boucle principale de Mean Shift
|
| 72 |
+
for iteration in range(max_iters):
|
| 73 |
+
# Calcul vectorisé des distances entre tous les pixels et les centres
|
| 74 |
+
distances = np.sqrt(np.sum((pixels[:, np.newaxis] - centers) ** 2, axis=2))
|
| 75 |
+
|
| 76 |
+
# Pour chaque centre, trouver les pixels dans son rayon et calculer la nouvelle position
|
| 77 |
+
new_centers = []
|
| 78 |
+
for i in range(len(centers)):
|
| 79 |
+
in_bandwidth = distances[:, i] < bandwidth
|
| 80 |
+
if np.sum(in_bandwidth) > 0:
|
| 81 |
+
new_centers.append(np.mean(pixels[in_bandwidth], axis=0))
|
| 82 |
+
else:
|
| 83 |
+
new_centers.append(centers[i])
|
| 84 |
+
|
| 85 |
+
new_centers = np.array(new_centers)
|
| 86 |
+
|
| 87 |
+
# Vérifier la convergence
|
| 88 |
+
center_shifts = np.sqrt(np.sum((centers - new_centers) ** 2, axis=1))
|
| 89 |
+
centers = new_centers
|
| 90 |
+
|
| 91 |
+
if np.all(center_shifts < tol):
|
| 92 |
+
print(f"Mean Shift convergé après {iteration + 1} itérations.")
|
| 93 |
+
break
|
| 94 |
+
|
| 95 |
+
# Attribution des labels
|
| 96 |
+
distances = np.sqrt(np.sum((pixels[:, np.newaxis] - centers) ** 2, axis=2))
|
| 97 |
+
labels = np.argmin(distances, axis=1)
|
| 98 |
+
|
| 99 |
+
# Génération des couleurs distinctes
|
| 100 |
+
n_clusters = len(centers)
|
| 101 |
+
distinct_colors = np.zeros((n_clusters, 3), dtype=np.uint8)
|
| 102 |
+
|
| 103 |
+
for i in range(n_clusters):
|
| 104 |
+
hue = i / n_clusters
|
| 105 |
+
saturation = 0.8
|
| 106 |
+
value = 0.8
|
| 107 |
+
rgb = colorsys.hsv_to_rgb(hue, saturation, value)
|
| 108 |
+
distinct_colors[i] = np.array(rgb) * 255
|
| 109 |
+
|
| 110 |
+
# Création de l'image segmentée
|
| 111 |
+
segmented_pixels = distinct_colors[labels]
|
| 112 |
+
segmented_image = segmented_pixels.reshape(image.shape)
|
| 113 |
+
|
| 114 |
+
# Redimensionnement au format original si nécessaire
|
| 115 |
+
if scale < 1:
|
| 116 |
+
segmented_image = cv2.resize(segmented_image, (w, h))
|
| 117 |
+
|
| 118 |
+
return segmented_image
|
| 119 |
+
|
| 120 |
+
def segment_image(image, k=None):
|
| 121 |
+
"""
|
| 122 |
+
Fonction principale qui gère la segmentation en fonction des paramètres
|
| 123 |
+
"""
|
| 124 |
+
if image is None:
|
| 125 |
+
return None
|
| 126 |
+
|
| 127 |
+
# Conversion de l'image au format BGR si nécessaire
|
| 128 |
+
if len(image.shape) == 2: # Image en niveaux de gris
|
| 129 |
+
image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
|
| 130 |
+
elif image.shape[2] == 4: # Image avec canal alpha
|
| 131 |
+
image = cv2.cvtColor(image, cv2.COLOR_RGBA2BGR)
|
| 132 |
+
|
| 133 |
+
try:
|
| 134 |
+
if k is not None and k > 0:
|
| 135 |
+
return k_means_segmentation(image, int(k))
|
| 136 |
+
else:
|
| 137 |
+
return mean_shift_segmentation(image)
|
| 138 |
+
except Exception as e:
|
| 139 |
+
print(f"Erreur lors de la segmentation: {str(e)}")
|
| 140 |
+
return None
|
| 141 |
+
|
| 142 |
+
|
| 143 |
+
def segment_image(image, k=None):
|
| 144 |
+
"""
|
| 145 |
+
Fonction principale qui gère la segmentation en fonction des paramètres
|
| 146 |
+
"""
|
| 147 |
+
if image is None:
|
| 148 |
+
return None
|
| 149 |
+
|
| 150 |
+
# Conversion de l'image au format BGR si nécessaire
|
| 151 |
+
if len(image.shape) == 2: # Image en niveaux de gris
|
| 152 |
+
image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
|
| 153 |
+
elif image.shape[2] == 4: # Image avec canal alpha
|
| 154 |
+
image = cv2.cvtColor(image, cv2.COLOR_RGBA2BGR)
|
| 155 |
+
|
| 156 |
+
try:
|
| 157 |
+
if k is not None and k > 0:
|
| 158 |
+
return k_means_segmentation(image, int(k))
|
| 159 |
+
else:
|
| 160 |
+
return mean_shift_segmentation(image)
|
| 161 |
+
except Exception as e:
|
| 162 |
+
print(f"Erreur lors de la segmentation: {str(e)}")
|
| 163 |
+
return None
|
| 164 |
+
|
| 165 |
+
|
| 166 |
+
with gr.Blocks(title="Art par Segmentation d'Images", theme=gr.themes.Soft()) as app:
|
| 167 |
+
gr.Markdown("""
|
| 168 |
+
# 🎨 Studio de Segmentation Artistique
|
| 169 |
+
|
| 170 |
+
Transformez vos photos en œuvres d'art segmentées !
|
| 171 |
+
|
| 172 |
+
### Mode d'emploi :
|
| 173 |
+
1. Téléchargez une image
|
| 174 |
+
2. Choisissez votre style :
|
| 175 |
+
- **K-means** : Entrez un nombre K pour définir le nombre exact de couleurs
|
| 176 |
+
- **Mean Shift** : Laissez K vide pour une segmentation automatique
|
| 177 |
+
3. Cliquez sur "Transformer" et admirez le résultat !
|
| 178 |
+
""")
|
| 179 |
+
|
| 180 |
+
with gr.Row():
|
| 181 |
+
with gr.Column(scale=1):
|
| 182 |
+
input_image = gr.Image(
|
| 183 |
+
label="Image Originale",
|
| 184 |
+
type="numpy"
|
| 185 |
+
)
|
| 186 |
+
k_input = gr.Number(
|
| 187 |
+
label="Nombre de segments (K)",
|
| 188 |
+
minimum=0,
|
| 189 |
+
step=1
|
| 190 |
+
)
|
| 191 |
+
segment_btn = gr.Button(
|
| 192 |
+
"🎨 Transformer",
|
| 193 |
+
variant="primary",
|
| 194 |
+
scale=0
|
| 195 |
+
)
|
| 196 |
+
|
| 197 |
+
with gr.Column(scale=1):
|
| 198 |
+
output_image = gr.Image(
|
| 199 |
+
label="Résultat",
|
| 200 |
+
type="numpy"
|
| 201 |
+
)
|
| 202 |
+
|
| 203 |
+
# Exemples d'utilisation
|
| 204 |
+
gr.Examples(
|
| 205 |
+
examples=[
|
| 206 |
+
["MEN-Denim-id_00000089-17_4_full.png", 5],
|
| 207 |
+
["ben.jpg", None, ""],
|
| 208 |
+
|
| 209 |
+
],
|
| 210 |
+
inputs=[input_image, k_input],
|
| 211 |
+
outputs=output_image,
|
| 212 |
+
fn=segment_image,
|
| 213 |
+
cache_examples=True
|
| 214 |
+
)
|
| 215 |
+
|
| 216 |
+
# Configuration des événements
|
| 217 |
+
segment_btn.click(
|
| 218 |
+
fn=segment_image,
|
| 219 |
+
inputs=[input_image, k_input],
|
| 220 |
+
outputs=output_image
|
| 221 |
+
)
|
| 222 |
+
|
| 223 |
+
# Message d'aide supplémentaire
|
| 224 |
+
gr.Markdown("""
|
| 225 |
+
### 💡 Conseils :
|
| 226 |
+
- Pour K-means : essayez des valeurs entre 3 et 10 pour des résultats intéressants
|
| 227 |
+
- Pour Mean Shift : idéal pour les images complexes avec beaucoup de détails
|
| 228 |
+
- Les images de taille moyenne fonctionnent le mieux
|
| 229 |
+
""")
|
| 230 |
+
|
| 231 |
+
# Lancement de l'application
|
| 232 |
+
app.launch(share=True)
|