Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -35,21 +35,29 @@ def ensure_model_exists():
|
|
| 35 |
|
| 36 |
ensure_model_exists()
|
| 37 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
# ============================================
|
| 39 |
-
# 2. دالة معالجة الصورة
|
| 40 |
# ============================================
|
| 41 |
def process_image_for_mnist(image_input):
|
|
|
|
| 42 |
if image_input is None:
|
| 43 |
return None, None
|
| 44 |
|
| 45 |
if isinstance(image_input, dict):
|
| 46 |
-
image = image_input.get('composite', None)
|
|
|
|
|
|
|
| 47 |
if image is None and len(image_input) > 0:
|
| 48 |
image = list(image_input.values())[0]
|
| 49 |
else:
|
| 50 |
image = image_input
|
| 51 |
|
| 52 |
-
if image is None or not isinstance(image, np.ndarray):
|
| 53 |
return None, None
|
| 54 |
|
| 55 |
if len(image.shape) == 3 and image.shape[-1] == 4:
|
|
@@ -71,7 +79,6 @@ def process_image_for_mnist(image_input):
|
|
| 71 |
|
| 72 |
return normalized, reshaped
|
| 73 |
|
| 74 |
-
# Variable لحفظ النموذج داخل الـ GPU Worker
|
| 75 |
loaded_model = None
|
| 76 |
|
| 77 |
# ============================================
|
|
@@ -81,18 +88,17 @@ loaded_model = None
|
|
| 81 |
def predict_sketch(image):
|
| 82 |
global loaded_model
|
| 83 |
try:
|
|
|
|
| 84 |
if image is None:
|
| 85 |
-
return {}, None, "⚠️ يرجى الرسم في المربع أولاً!"
|
| 86 |
|
| 87 |
normalized, reshaped = process_image_for_mnist(image)
|
| 88 |
|
| 89 |
if normalized is None:
|
| 90 |
return {}, None, "⚠️ تعذر معالجة الصورة المدخلة."
|
| 91 |
|
| 92 |
-
# تحميل TensorFlow وتكوين الـ GPU ديناميكياً داخل بيئة ZeroGPU فقط
|
| 93 |
import tensorflow as tf
|
| 94 |
|
| 95 |
-
# تهيئة نمو الذاكرة لتفادي انهيار CUDA memory allocation
|
| 96 |
gpus = tf.config.list_physical_devices('GPU')
|
| 97 |
if gpus:
|
| 98 |
try:
|
|
@@ -104,7 +110,6 @@ def predict_sketch(image):
|
|
| 104 |
if loaded_model is None:
|
| 105 |
loaded_model = tf.keras.models.load_model(model_path)
|
| 106 |
|
| 107 |
-
# التوقع
|
| 108 |
preds = loaded_model(reshaped, training=False).numpy()[0]
|
| 109 |
predicted_class = int(np.argmax(preds))
|
| 110 |
confidence = float(np.max(preds))
|
|
@@ -130,7 +135,17 @@ def predict_sketch(image):
|
|
| 130 |
return {}, None, f"❌ خطأ أثناء التوقّع: {str(e)}"
|
| 131 |
|
| 132 |
# ============================================
|
| 133 |
-
# 4.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
# ============================================
|
| 135 |
with gr.Blocks(title="MNIST Recognizer on ZeroGPU") as demo:
|
| 136 |
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد (ZeroGPU)")
|
|
@@ -138,17 +153,26 @@ with gr.Blocks(title="MNIST Recognizer on ZeroGPU") as demo:
|
|
| 138 |
with gr.Row():
|
| 139 |
with gr.Column(scale=1):
|
| 140 |
sketch = gr.Sketchpad(label="✏️ منطقة الرسم")
|
| 141 |
-
|
|
|
|
|
|
|
| 142 |
info = gr.Textbox(label="📌 معلومات التصحيح", interactive=False, lines=8)
|
| 143 |
|
| 144 |
with gr.Column(scale=1):
|
| 145 |
output = gr.Label(num_top_classes=3, label="📊 احتمالات التوقع")
|
| 146 |
processed_img = gr.Image(label="🖼️ الصورة بعد المعالجة", image_mode="L")
|
| 147 |
|
|
|
|
| 148 |
submit_btn.click(
|
| 149 |
fn=predict_sketch,
|
| 150 |
inputs=sketch,
|
| 151 |
outputs=[output, processed_img, info]
|
| 152 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
|
| 154 |
demo.launch()
|
|
|
|
| 35 |
|
| 36 |
ensure_model_exists()
|
| 37 |
|
| 38 |
+
# تحميل بيانات الاختبار للأمثلة العشوائية
|
| 39 |
+
import tensorflow as tf
|
| 40 |
+
(_, _), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()
|
| 41 |
+
test_images_norm = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255.0
|
| 42 |
+
|
| 43 |
# ============================================
|
| 44 |
+
# 2. دالة معالجة الصورة الآمنة
|
| 45 |
# ============================================
|
| 46 |
def process_image_for_mnist(image_input):
|
| 47 |
+
# التحقق الآمن من الصفر/الغياب بدون تقييم المصفوفة كـ Boolean
|
| 48 |
if image_input is None:
|
| 49 |
return None, None
|
| 50 |
|
| 51 |
if isinstance(image_input, dict):
|
| 52 |
+
image = image_input.get('composite', None)
|
| 53 |
+
if image is None:
|
| 54 |
+
image = image_input.get('background', None)
|
| 55 |
if image is None and len(image_input) > 0:
|
| 56 |
image = list(image_input.values())[0]
|
| 57 |
else:
|
| 58 |
image = image_input
|
| 59 |
|
| 60 |
+
if image is None or not isinstance(image, np.ndarray) or image.size == 0:
|
| 61 |
return None, None
|
| 62 |
|
| 63 |
if len(image.shape) == 3 and image.shape[-1] == 4:
|
|
|
|
| 79 |
|
| 80 |
return normalized, reshaped
|
| 81 |
|
|
|
|
| 82 |
loaded_model = None
|
| 83 |
|
| 84 |
# ============================================
|
|
|
|
| 88 |
def predict_sketch(image):
|
| 89 |
global loaded_model
|
| 90 |
try:
|
| 91 |
+
# إصلاح سبب الخطأ: استخدام size أو is None بشكل صريح
|
| 92 |
if image is None:
|
| 93 |
+
return {}, None, "⚠️ يرجى الرسم في المربع أو اختيار مثال أولاً!"
|
| 94 |
|
| 95 |
normalized, reshaped = process_image_for_mnist(image)
|
| 96 |
|
| 97 |
if normalized is None:
|
| 98 |
return {}, None, "⚠️ تعذر معالجة الصورة المدخلة."
|
| 99 |
|
|
|
|
| 100 |
import tensorflow as tf
|
| 101 |
|
|
|
|
| 102 |
gpus = tf.config.list_physical_devices('GPU')
|
| 103 |
if gpus:
|
| 104 |
try:
|
|
|
|
| 110 |
if loaded_model is None:
|
| 111 |
loaded_model = tf.keras.models.load_model(model_path)
|
| 112 |
|
|
|
|
| 113 |
preds = loaded_model(reshaped, training=False).numpy()[0]
|
| 114 |
predicted_class = int(np.argmax(preds))
|
| 115 |
confidence = float(np.max(preds))
|
|
|
|
| 135 |
return {}, None, f"❌ خطأ أثناء التوقّع: {str(e)}"
|
| 136 |
|
| 137 |
# ============================================
|
| 138 |
+
# 4. دالة اختيار مثال عشوائي (تمت إعادتها)
|
| 139 |
+
# ============================================
|
| 140 |
+
def random_example():
|
| 141 |
+
idx = np.random.randint(0, len(test_images))
|
| 142 |
+
img = test_images[idx]
|
| 143 |
+
img_large = cv2.resize(img, (280, 280), interpolation=cv2.INTER_NEAREST)
|
| 144 |
+
img_rgb = np.stack([img_large] * 3, axis=2)
|
| 145 |
+
return img_rgb, f"🎲 تم اختيار رقم عشوائي (الرقم الحقيقي: {test_labels[idx]})"
|
| 146 |
+
|
| 147 |
+
# ============================================
|
| 148 |
+
# 5. بناء واجهة Gradio
|
| 149 |
# ============================================
|
| 150 |
with gr.Blocks(title="MNIST Recognizer on ZeroGPU") as demo:
|
| 151 |
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد (ZeroGPU)")
|
|
|
|
| 153 |
with gr.Row():
|
| 154 |
with gr.Column(scale=1):
|
| 155 |
sketch = gr.Sketchpad(label="✏️ منطقة الرسم")
|
| 156 |
+
with gr.Row():
|
| 157 |
+
submit_btn = gr.Button("🔮 توقع", variant="primary")
|
| 158 |
+
random_btn = gr.Button("🎲 مثال عشوائي", variant="secondary")
|
| 159 |
info = gr.Textbox(label="📌 معلومات التصحيح", interactive=False, lines=8)
|
| 160 |
|
| 161 |
with gr.Column(scale=1):
|
| 162 |
output = gr.Label(num_top_classes=3, label="📊 احتمالات التوقع")
|
| 163 |
processed_img = gr.Image(label="🖼️ الصورة بعد المعالجة", image_mode="L")
|
| 164 |
|
| 165 |
+
# ربط الأحداث
|
| 166 |
submit_btn.click(
|
| 167 |
fn=predict_sketch,
|
| 168 |
inputs=sketch,
|
| 169 |
outputs=[output, processed_img, info]
|
| 170 |
)
|
| 171 |
+
|
| 172 |
+
random_btn.click(
|
| 173 |
+
fn=random_example,
|
| 174 |
+
inputs=[],
|
| 175 |
+
outputs=[sketch, info]
|
| 176 |
+
)
|
| 177 |
|
| 178 |
demo.launch()
|