Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,10 +4,10 @@ import tensorflow as tf
|
|
| 4 |
import cv2
|
| 5 |
from tensorflow.keras import datasets, layers, models
|
| 6 |
import os
|
| 7 |
-
import spaces
|
| 8 |
|
| 9 |
# ============================================
|
| 10 |
-
# 1. تحميل
|
| 11 |
# ============================================
|
| 12 |
model_path = 'mnist_cnn_model.keras'
|
| 13 |
|
|
@@ -42,24 +42,50 @@ else:
|
|
| 42 |
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
|
| 43 |
|
| 44 |
# ============================================
|
| 45 |
-
# 3. د
|
| 46 |
# ============================================
|
| 47 |
-
# --- الدالة التي تستخدم GPU يتم تزيينها بـ @spaces.GPU ---
|
| 48 |
@spaces.GPU
|
| 49 |
def predict_sketch(image):
|
| 50 |
try:
|
| 51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
resized = cv2.resize(gray, (28, 28))
|
| 53 |
-
inverted = 255 - resized
|
| 54 |
-
normalized = inverted.astype('float32') / 255.0
|
| 55 |
-
reshaped = normalized.reshape(1, 28, 28, 1)
|
| 56 |
|
| 57 |
-
|
| 58 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
except Exception as e:
|
| 60 |
return {"خطأ": str(e)}
|
| 61 |
|
| 62 |
-
#
|
|
|
|
|
|
|
| 63 |
def random_example():
|
| 64 |
idx = np.random.randint(0, len(test_images))
|
| 65 |
img = test_images[idx].reshape(28, 28)
|
|
@@ -67,7 +93,7 @@ def random_example():
|
|
| 67 |
return img_rgb, f"الرقم الحقيقي: {test_labels[idx]}"
|
| 68 |
|
| 69 |
# ============================================
|
| 70 |
-
#
|
| 71 |
# ============================================
|
| 72 |
with gr.Blocks(title="MNIST Recognizer") as demo:
|
| 73 |
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد")
|
|
@@ -88,6 +114,6 @@ with gr.Blocks(title="MNIST Recognizer") as demo:
|
|
| 88 |
random_btn.click(fn=random_example, inputs=[], outputs=[sketch, info])
|
| 89 |
|
| 90 |
# ============================================
|
| 91 |
-
#
|
| 92 |
# ============================================
|
| 93 |
demo.launch()
|
|
|
|
| 4 |
import cv2
|
| 5 |
from tensorflow.keras import datasets, layers, models
|
| 6 |
import os
|
| 7 |
+
import spaces
|
| 8 |
|
| 9 |
# ============================================
|
| 10 |
+
# 1. تحميل أو بناء النموذج
|
| 11 |
# ============================================
|
| 12 |
model_path = 'mnist_cnn_model.keras'
|
| 13 |
|
|
|
|
| 42 |
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
|
| 43 |
|
| 44 |
# ============================================
|
| 45 |
+
# 3. دالة معالجة الصورة والتنبؤ (ذكية)
|
| 46 |
# ============================================
|
|
|
|
| 47 |
@spaces.GPU
|
| 48 |
def predict_sketch(image):
|
| 49 |
try:
|
| 50 |
+
# 1. تحويل إلى تدرج رمادي
|
| 51 |
+
if len(image.shape) == 3:
|
| 52 |
+
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
| 53 |
+
else:
|
| 54 |
+
gray = image
|
| 55 |
+
|
| 56 |
+
# 2. تغيير الحجم إلى 28x28
|
| 57 |
resized = cv2.resize(gray, (28, 28))
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
+
# 3. تطبيع القيم إلى [0,1]
|
| 60 |
+
normalized = resized.astype('float32') / 255.0
|
| 61 |
+
|
| 62 |
+
# 4. **المعالجة الذكية: نحاول كلا الاتجاهين ونأخذ الأفضل**
|
| 63 |
+
# MNIST تدربت على خلفية سوداء (0) وأرقام بيضاء (1)
|
| 64 |
+
# لكن Sketchpad قد يعطي خلفية بيضاء وأرقام سوداء
|
| 65 |
+
|
| 66 |
+
# الخيار 1: الصورة كما هي (افتراض أن الخلفية سوداء والرقم أبيض)
|
| 67 |
+
img1 = normalized.reshape(1, 28, 28, 1)
|
| 68 |
+
pred1 = model.predict(img1, verbose=0)[0]
|
| 69 |
+
|
| 70 |
+
# الخيار 2: قلب الألوان (افتراض أن الخلفية بيضاء والرقم أسود)
|
| 71 |
+
img2 = (1 - normalized).reshape(1, 28, 28, 1)
|
| 72 |
+
pred2 = model.predict(img2, verbose=0)[0]
|
| 73 |
+
|
| 74 |
+
# اختيار النتيجة ذات الثقة الأعلى
|
| 75 |
+
max_conf1 = np.max(pred1)
|
| 76 |
+
max_conf2 = np.max(pred2)
|
| 77 |
+
|
| 78 |
+
if max_conf1 >= max_conf2:
|
| 79 |
+
return {str(i): float(pred1[i]) for i in range(10)}
|
| 80 |
+
else:
|
| 81 |
+
return {str(i): float(pred2[i]) for i in range(10)}
|
| 82 |
+
|
| 83 |
except Exception as e:
|
| 84 |
return {"خطأ": str(e)}
|
| 85 |
|
| 86 |
+
# ============================================
|
| 87 |
+
# 4. دالة المثال العشوائي
|
| 88 |
+
# ============================================
|
| 89 |
def random_example():
|
| 90 |
idx = np.random.randint(0, len(test_images))
|
| 91 |
img = test_images[idx].reshape(28, 28)
|
|
|
|
| 93 |
return img_rgb, f"الرقم الحقيقي: {test_labels[idx]}"
|
| 94 |
|
| 95 |
# ============================================
|
| 96 |
+
# 5. بناء واجهة Gradio
|
| 97 |
# ============================================
|
| 98 |
with gr.Blocks(title="MNIST Recognizer") as demo:
|
| 99 |
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد")
|
|
|
|
| 114 |
random_btn.click(fn=random_example, inputs=[], outputs=[sketch, info])
|
| 115 |
|
| 116 |
# ============================================
|
| 117 |
+
# 6. تشغيل التطبيق (في النطاق العام لـ ZeroGPU)
|
| 118 |
# ============================================
|
| 119 |
demo.launch()
|