Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,16 +5,11 @@ import cv2
|
|
| 5 |
from tensorflow.keras import datasets, layers, models
|
| 6 |
import os
|
| 7 |
|
| 8 |
-
# ====================
|
| 9 |
-
# 1. تحميل أو بناء النموذج (حل احتياطي في حال عدم وجود الملف)
|
| 10 |
-
# ============================================================
|
| 11 |
model_path = 'mnist_cnn_model.keras'
|
| 12 |
-
|
| 13 |
if os.path.exists(model_path):
|
| 14 |
model = tf.keras.models.load_model(model_path)
|
| 15 |
-
print("✅ تم تحميل النموذج المحفوظ.")
|
| 16 |
else:
|
| 17 |
-
print("⚠️ لم يتم العثور على النموذج، جارٍ البناء والتدريب...")
|
| 18 |
model = models.Sequential([
|
| 19 |
layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
|
| 20 |
layers.MaxPooling2D((2, 2)),
|
|
@@ -24,80 +19,49 @@ else:
|
|
| 24 |
layers.Dense(64, activation='relu'),
|
| 25 |
layers.Dense(10, activation='softmax')
|
| 26 |
])
|
| 27 |
-
model.compile(optimizer='adam',
|
| 28 |
-
loss='sparse_categorical_crossentropy',
|
| 29 |
-
metrics=['accuracy'])
|
| 30 |
-
|
| 31 |
(train_images, train_labels), _ = datasets.mnist.load_data()
|
| 32 |
train_images = train_images.reshape((60000, 28, 28, 1)).astype('float32') / 255
|
| 33 |
model.fit(train_images, train_labels, epochs=3, validation_split=0.1, verbose=1)
|
| 34 |
model.save(model_path)
|
| 35 |
-
print("✅ تم بناء النموذج وتدريبه وحفظه.")
|
| 36 |
|
| 37 |
-
# ====================
|
| 38 |
-
# 2. تحميل بيانات MNIST للاستخدام في الأمثلة العشوائية
|
| 39 |
-
# ============================================================
|
| 40 |
(_, _), (test_images, test_labels) = datasets.mnist.load_data()
|
| 41 |
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
|
| 42 |
|
| 43 |
-
# ====================
|
| 44 |
-
# 3. دالة التنبؤ بالصورة المرسومة
|
| 45 |
-
# ============================================================
|
| 46 |
def predict_sketch(image):
|
| 47 |
-
"""
|
| 48 |
-
تستقبل الصورة من Sketchpad، تعالجها، وتعيد الاحتمالات لكل فئة.
|
| 49 |
-
"""
|
| 50 |
try:
|
| 51 |
-
# الصورة قادمة بصيغة (height, width, 3) من نوع numpy array
|
| 52 |
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
| 53 |
resized = cv2.resize(gray, (28, 28))
|
| 54 |
-
# قلب الألوان لأن MNIST تدربت على خلفية سوداء وكتابة بيضاء
|
| 55 |
inverted = 255 - resized
|
| 56 |
normalized = inverted.astype('float32') / 255.0
|
| 57 |
reshaped = normalized.reshape(1, 28, 28, 1)
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
# تحويل إلى قاموس لعرضه في Label
|
| 61 |
-
return {str(i): float(prediction[i]) for i in range(10)}
|
| 62 |
except Exception as e:
|
| 63 |
-
return {"
|
| 64 |
|
| 65 |
-
# ============================================================
|
| 66 |
-
# 4. دالة جلب مثال عشوائي من بيانات الاختبار
|
| 67 |
-
# ============================================================
|
| 68 |
def random_example():
|
| 69 |
idx = np.random.randint(0, len(test_images))
|
| 70 |
img = test_images[idx].reshape(28, 28)
|
| 71 |
-
# تحويل الصورة إلى 3 قنوات (RGB) لأن Sketchpad يتوقع ذلك
|
| 72 |
img_rgb = np.stack([img] * 3, axis=2)
|
| 73 |
-
|
| 74 |
-
return img_rgb, info_text
|
| 75 |
-
|
| 76 |
-
# ============================================================
|
| 77 |
-
# 5. بناء واجهة Gradio باستخدام Blocks + Sketchpad
|
| 78 |
-
# ============================================================
|
| 79 |
-
with gr.Blocks(title="MNIST Digit Recognizer") as demo:
|
| 80 |
-
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد")
|
| 81 |
-
gr.Markdown("ارسم رقماً (0-9) في المربع، أو اضغط على زر **'مثال عشوائي'** لتجربة النموذج.")
|
| 82 |
|
|
|
|
|
|
|
|
|
|
| 83 |
with gr.Row():
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
sketch = gr.Sketchpad(label="✏️ ارسم هنا")
|
| 87 |
with gr.Row():
|
| 88 |
-
submit_btn = gr.Button("
|
| 89 |
-
random_btn = gr.Button("
|
| 90 |
-
info = gr.Textbox(label="
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
with gr.Column(scale=1):
|
| 94 |
-
output = gr.Label(num_top_classes=3, label="📊 أعلى 3 احتمالات")
|
| 95 |
-
|
| 96 |
-
# ربط الأزرار بالدوال
|
| 97 |
submit_btn.click(fn=predict_sketch, inputs=sketch, outputs=output)
|
| 98 |
random_btn.click(fn=random_example, inputs=[], outputs=[sketch, info])
|
| 99 |
|
| 100 |
-
# ====================
|
| 101 |
-
#
|
| 102 |
-
demo.launch()
|
| 103 |
-
# ============================================================
|
|
|
|
| 5 |
from tensorflow.keras import datasets, layers, models
|
| 6 |
import os
|
| 7 |
|
| 8 |
+
# ========== 1. تحميل النموذج ==========
|
|
|
|
|
|
|
| 9 |
model_path = 'mnist_cnn_model.keras'
|
|
|
|
| 10 |
if os.path.exists(model_path):
|
| 11 |
model = tf.keras.models.load_model(model_path)
|
|
|
|
| 12 |
else:
|
|
|
|
| 13 |
model = models.Sequential([
|
| 14 |
layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
|
| 15 |
layers.MaxPooling2D((2, 2)),
|
|
|
|
| 19 |
layers.Dense(64, activation='relu'),
|
| 20 |
layers.Dense(10, activation='softmax')
|
| 21 |
])
|
| 22 |
+
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
|
|
|
|
|
|
|
|
|
|
| 23 |
(train_images, train_labels), _ = datasets.mnist.load_data()
|
| 24 |
train_images = train_images.reshape((60000, 28, 28, 1)).astype('float32') / 255
|
| 25 |
model.fit(train_images, train_labels, epochs=3, validation_split=0.1, verbose=1)
|
| 26 |
model.save(model_path)
|
|
|
|
| 27 |
|
| 28 |
+
# ========== 2. بيانات MNIST ==========
|
|
|
|
|
|
|
| 29 |
(_, _), (test_images, test_labels) = datasets.mnist.load_data()
|
| 30 |
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
|
| 31 |
|
| 32 |
+
# ========== 3. دوال التنبؤ والعشوائي ==========
|
|
|
|
|
|
|
| 33 |
def predict_sketch(image):
|
|
|
|
|
|
|
|
|
|
| 34 |
try:
|
|
|
|
| 35 |
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
| 36 |
resized = cv2.resize(gray, (28, 28))
|
|
|
|
| 37 |
inverted = 255 - resized
|
| 38 |
normalized = inverted.astype('float32') / 255.0
|
| 39 |
reshaped = normalized.reshape(1, 28, 28, 1)
|
| 40 |
+
pred = model.predict(reshaped, verbose=0)[0]
|
| 41 |
+
return {str(i): float(pred[i]) for i in range(10)}
|
|
|
|
|
|
|
| 42 |
except Exception as e:
|
| 43 |
+
return {"error": str(e)}
|
| 44 |
|
|
|
|
|
|
|
|
|
|
| 45 |
def random_example():
|
| 46 |
idx = np.random.randint(0, len(test_images))
|
| 47 |
img = test_images[idx].reshape(28, 28)
|
|
|
|
| 48 |
img_rgb = np.stack([img] * 3, axis=2)
|
| 49 |
+
return img_rgb, f"الرقم الحقيقي: {test_labels[idx]}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
+
# ========== 4. الواجهة (بدون launch) ==========
|
| 52 |
+
with gr.Blocks(title="MNIST Recognizer") as demo:
|
| 53 |
+
gr.Markdown("# 🧠 التعرف على الأرقام")
|
| 54 |
with gr.Row():
|
| 55 |
+
with gr.Column():
|
| 56 |
+
sketch = gr.Sketchpad(label="ارسم هنا")
|
|
|
|
| 57 |
with gr.Row():
|
| 58 |
+
submit_btn = gr.Button("توقع", variant="primary")
|
| 59 |
+
random_btn = gr.Button("عشوائي", variant="secondary")
|
| 60 |
+
info = gr.Textbox(label="معلومات")
|
| 61 |
+
with gr.Column():
|
| 62 |
+
output = gr.Label(num_top_classes=3, label="الاحتمالات")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
submit_btn.click(fn=predict_sketch, inputs=sketch, outputs=output)
|
| 64 |
random_btn.click(fn=random_example, inputs=[], outputs=[sketch, info])
|
| 65 |
|
| 66 |
+
# ========== 5. ❌ ممنوع استخدام launch() هنا ==========
|
| 67 |
+
# Spaces يتولى التشغيل عبر المتغير 'demo'
|
|
|
|
|
|