Spaces:
Build error
Build error
metadata v12
Browse files
app.py
CHANGED
@@ -8,25 +8,24 @@ import pickle
|
|
8 |
from pathlib import Path
|
9 |
import gc
|
10 |
|
11 |
-
#
|
12 |
-
HF_TOKEN = os.getenv("HF_TOKEN")
|
13 |
-
|
14 |
-
# 📁 Directorio para embeddings
|
15 |
EMBEDDINGS_DIR = Path("embeddings")
|
16 |
EMBEDDINGS_DIR.mkdir(exist_ok=True)
|
17 |
EMBEDDINGS_FILE = EMBEDDINGS_DIR / "embeddings.pkl"
|
18 |
|
19 |
-
# ✅ Cargar dataset
|
20 |
dataset = load_dataset(
|
21 |
"csv",
|
22 |
data_files="metadata.csv",
|
23 |
-
split="train"
|
24 |
-
)
|
|
|
25 |
print("✅ Primer item:", dataset[0])
|
26 |
|
|
|
27 |
dataset = dataset.cast_column("image", HfImage())
|
28 |
|
29 |
-
# 🔄 Preprocesar imagen para
|
30 |
def preprocess_image(img: Image.Image) -> np.ndarray:
|
31 |
img_rgb = img.convert("RGB")
|
32 |
img_resized = img_rgb.resize((160, 160), Image.Resampling.LANCZOS)
|
@@ -35,11 +34,11 @@ def preprocess_image(img: Image.Image) -> np.ndarray:
|
|
35 |
# 📦 Construir base de datos de embeddings
|
36 |
def build_database():
|
37 |
if EMBEDDINGS_FILE.exists():
|
38 |
-
print("📂 Cargando embeddings desde
|
39 |
-
with open(EMBEDDINGS_FILE,
|
40 |
return pickle.load(f)
|
41 |
|
42 |
-
print("🔄 Calculando embeddings
|
43 |
database = []
|
44 |
batch_size = 10
|
45 |
|
@@ -47,15 +46,10 @@ def build_database():
|
|
47 |
batch = dataset[i:i + batch_size]
|
48 |
print(f"📦 Procesando lote {i // batch_size + 1}/{(len(dataset) + batch_size - 1) // batch_size}")
|
49 |
|
50 |
-
for j,
|
51 |
try:
|
52 |
-
if not isinstance(item, dict) or "image" not in item:
|
53 |
-
print(f"⚠️ Saltando item {i+j} - estructura inválida: {item}")
|
54 |
-
continue
|
55 |
-
|
56 |
-
img = item["image"]
|
57 |
if not isinstance(img, Image.Image):
|
58 |
-
print(f"⚠️ Saltando item {i+j} - no es imagen: {type(img)}")
|
59 |
continue
|
60 |
|
61 |
img_processed = preprocess_image(img)
|
@@ -65,20 +59,20 @@ def build_database():
|
|
65 |
enforce_detection=False
|
66 |
)[0]["embedding"]
|
67 |
|
68 |
-
database.append((f"image_{i+j}", img, embedding))
|
69 |
-
print(f"✅ Procesada imagen {i+j+1}/{len(dataset)}")
|
70 |
|
71 |
del img_processed
|
72 |
gc.collect()
|
73 |
|
74 |
except Exception as e:
|
75 |
-
print(f"❌ Error al procesar imagen {i+j}: {str(e)}")
|
76 |
continue
|
77 |
|
78 |
-
# Guardar después de cada
|
79 |
if database:
|
80 |
-
print("💾 Guardando
|
81 |
-
with open(EMBEDDINGS_FILE,
|
82 |
pickle.dump(database, f)
|
83 |
|
84 |
gc.collect()
|
@@ -97,8 +91,8 @@ def find_similar_faces(uploaded_image: Image.Image):
|
|
97 |
del img_processed
|
98 |
gc.collect()
|
99 |
except Exception as e:
|
100 |
-
print(f"Error al procesar imagen de
|
101 |
-
return [], "⚠ No se detectó un rostro válido
|
102 |
|
103 |
similarities = []
|
104 |
for name, db_img, embedding in database:
|
@@ -110,18 +104,18 @@ def find_similar_faces(uploaded_image: Image.Image):
|
|
110 |
top_matches = similarities[:5]
|
111 |
|
112 |
gallery_items = []
|
113 |
-
|
114 |
for sim, name, img in top_matches:
|
115 |
caption = f"{name} - Similitud: {sim:.2f}"
|
116 |
gallery_items.append((img, caption))
|
117 |
-
|
118 |
|
119 |
-
return gallery_items,
|
120 |
|
121 |
-
#
|
122 |
print("🚀 Iniciando aplicación...")
|
123 |
database = build_database()
|
124 |
-
print(f"✅ Base
|
125 |
|
126 |
# 🎛️ Interfaz Gradio
|
127 |
demo = gr.Interface(
|
@@ -129,10 +123,10 @@ demo = gr.Interface(
|
|
129 |
inputs=gr.Image(label="📤 Sube una imagen", type="pil"),
|
130 |
outputs=[
|
131 |
gr.Gallery(label="📸 Rostros más similares"),
|
132 |
-
gr.Textbox(label="🧠
|
133 |
],
|
134 |
title="🔍 Buscador de Rostros con DeepFace",
|
135 |
-
description="Sube una imagen y se comparará contra los rostros del dataset
|
136 |
)
|
137 |
|
138 |
demo.launch()
|
|
|
8 |
from pathlib import Path
|
9 |
import gc
|
10 |
|
11 |
+
# 📁 Directorio para almacenar embeddings
|
|
|
|
|
|
|
12 |
EMBEDDINGS_DIR = Path("embeddings")
|
13 |
EMBEDDINGS_DIR.mkdir(exist_ok=True)
|
14 |
EMBEDDINGS_FILE = EMBEDDINGS_DIR / "embeddings.pkl"
|
15 |
|
16 |
+
# ✅ Cargar dataset desde metadata.csv (con URLs absolutas)
|
17 |
dataset = load_dataset(
|
18 |
"csv",
|
19 |
data_files="metadata.csv",
|
20 |
+
split="train"
|
21 |
+
)
|
22 |
+
|
23 |
print("✅ Primer item:", dataset[0])
|
24 |
|
25 |
+
# 🖼️ Convertir columna a imágenes usando HfImage (PIL)
|
26 |
dataset = dataset.cast_column("image", HfImage())
|
27 |
|
28 |
+
# 🔄 Preprocesar imagen para DeepFace
|
29 |
def preprocess_image(img: Image.Image) -> np.ndarray:
|
30 |
img_rgb = img.convert("RGB")
|
31 |
img_resized = img_rgb.resize((160, 160), Image.Resampling.LANCZOS)
|
|
|
34 |
# 📦 Construir base de datos de embeddings
|
35 |
def build_database():
|
36 |
if EMBEDDINGS_FILE.exists():
|
37 |
+
print("📂 Cargando embeddings desde archivo...")
|
38 |
+
with open(EMBEDDINGS_FILE, "rb") as f:
|
39 |
return pickle.load(f)
|
40 |
|
41 |
+
print("🔄 Calculando embeddings...")
|
42 |
database = []
|
43 |
batch_size = 10
|
44 |
|
|
|
46 |
batch = dataset[i:i + batch_size]
|
47 |
print(f"📦 Procesando lote {i // batch_size + 1}/{(len(dataset) + batch_size - 1) // batch_size}")
|
48 |
|
49 |
+
for j, img in enumerate(batch):
|
50 |
try:
|
|
|
|
|
|
|
|
|
|
|
51 |
if not isinstance(img, Image.Image):
|
52 |
+
print(f"⚠️ Saltando item {i + j} - no es imagen: {type(img)}")
|
53 |
continue
|
54 |
|
55 |
img_processed = preprocess_image(img)
|
|
|
59 |
enforce_detection=False
|
60 |
)[0]["embedding"]
|
61 |
|
62 |
+
database.append((f"image_{i + j}", img, embedding))
|
63 |
+
print(f"✅ Procesada imagen {i + j + 1}/{len(dataset)}")
|
64 |
|
65 |
del img_processed
|
66 |
gc.collect()
|
67 |
|
68 |
except Exception as e:
|
69 |
+
print(f"❌ Error al procesar imagen {i + j}: {str(e)}")
|
70 |
continue
|
71 |
|
72 |
+
# Guardar después de cada batch
|
73 |
if database:
|
74 |
+
print("💾 Guardando embeddings...")
|
75 |
+
with open(EMBEDDINGS_FILE, "wb") as f:
|
76 |
pickle.dump(database, f)
|
77 |
|
78 |
gc.collect()
|
|
|
91 |
del img_processed
|
92 |
gc.collect()
|
93 |
except Exception as e:
|
94 |
+
print(f"Error al procesar imagen de entrada: {str(e)}")
|
95 |
+
return [], "⚠ No se detectó un rostro válido."
|
96 |
|
97 |
similarities = []
|
98 |
for name, db_img, embedding in database:
|
|
|
104 |
top_matches = similarities[:5]
|
105 |
|
106 |
gallery_items = []
|
107 |
+
summary = ""
|
108 |
for sim, name, img in top_matches:
|
109 |
caption = f"{name} - Similitud: {sim:.2f}"
|
110 |
gallery_items.append((img, caption))
|
111 |
+
summary += caption + "\n"
|
112 |
|
113 |
+
return gallery_items, summary
|
114 |
|
115 |
+
# 🚀 Inicializar app
|
116 |
print("🚀 Iniciando aplicación...")
|
117 |
database = build_database()
|
118 |
+
print(f"✅ Base cargada con {len(database)} imágenes.")
|
119 |
|
120 |
# 🎛️ Interfaz Gradio
|
121 |
demo = gr.Interface(
|
|
|
123 |
inputs=gr.Image(label="📤 Sube una imagen", type="pil"),
|
124 |
outputs=[
|
125 |
gr.Gallery(label="📸 Rostros más similares"),
|
126 |
+
gr.Textbox(label="🧠 Resumen de similitud", lines=6)
|
127 |
],
|
128 |
title="🔍 Buscador de Rostros con DeepFace",
|
129 |
+
description="Sube una imagen y se comparará contra los rostros del dataset `Segizu/facial-recognition`."
|
130 |
)
|
131 |
|
132 |
demo.launch()
|