Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,146 +1,88 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import cv2
|
| 3 |
import gradio as gr
|
| 4 |
import numpy as np
|
| 5 |
import tensorflow as tf
|
| 6 |
-
|
| 7 |
-
from tensorflow.keras import layers, models
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
@spaces.GPU
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
optimizer="adam",
|
| 42 |
-
loss="sparse_categorical_crossentropy",
|
| 43 |
-
metrics=["accuracy"],
|
| 44 |
-
)
|
| 45 |
-
|
| 46 |
-
# تحميل البيانات وتقسيمها يدوياً لتفادي أخطاء Keras 3
|
| 47 |
-
(x_train_full, y_train_full), _ = tf.keras.datasets.mnist.load_data()
|
| 48 |
-
x_train_full = x_train_full.reshape(-1, 28, 28, 1).astype("float32") / 255.0
|
| 49 |
-
|
| 50 |
-
split = int(len(x_train_full) * 0.9)
|
| 51 |
-
x_tr, x_va = x_train_full[:split], x_train_full[split:]
|
| 52 |
-
y_tr, y_va = y_train_full[:split], y_train_full[split:]
|
| 53 |
-
|
| 54 |
-
# تدريب خفيف جداً لـ 3 جولات فقط (سريع ومناسب لـ CPU Space)
|
| 55 |
-
model.fit(
|
| 56 |
-
x_tr,
|
| 57 |
-
y_tr,
|
| 58 |
-
epochs=3,
|
| 59 |
-
batch_size=128,
|
| 60 |
-
validation_data=(x_va, y_va),
|
| 61 |
-
verbose=1,
|
| 62 |
-
)
|
| 63 |
-
|
| 64 |
-
model.save(MODEL_PATH)
|
| 65 |
-
print("✅ تم تدريب النموذج وحفظه بنجاح")
|
| 66 |
-
return model
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
# تحميل أو بناء النموذج
|
| 70 |
-
model = get_or_build_model()
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
# دالة تمركز كتلة الرسم
|
| 74 |
-
def center_image(img):
|
| 75 |
-
cy, cx = ndimage.center_of_mass(img)
|
| 76 |
-
rows, cols = img.shape
|
| 77 |
-
shiftx = np.round(cols / 2.0 - cx)
|
| 78 |
-
shifty = np.round(rows / 2.0 - cy)
|
| 79 |
-
M = np.float32([[1, 0, shiftx], [0, 1, shifty]])
|
| 80 |
-
return cv2.warpAffine(img, M, (cols, rows))
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
# دالة التنبؤ
|
| 84 |
-
def predict(image):
|
| 85 |
-
if image is None:
|
| 86 |
-
return None, {"error": "لم يتم الرسم"}
|
| 87 |
-
|
| 88 |
-
if isinstance(image, dict):
|
| 89 |
-
image = image.get("composite", image.get("background"))
|
| 90 |
-
|
| 91 |
-
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
| 92 |
-
_, thresh = cv2.threshold(gray, 200, 255, cv2.THRESH_BINARY_INV)
|
| 93 |
-
|
| 94 |
-
coords = cv2.findNonZero(thresh)
|
| 95 |
-
if coords is not None:
|
| 96 |
-
x, y, w, h = cv2.boundingRect(coords)
|
| 97 |
-
cropped = thresh[y : y + h, x : x + w]
|
| 98 |
-
|
| 99 |
-
if w > h:
|
| 100 |
-
new_w, new_h = 20, int(h * (20 / w))
|
| 101 |
-
else:
|
| 102 |
-
new_h, new_w = 20, int(w * (20 / h))
|
| 103 |
-
|
| 104 |
-
resized = cv2.resize(
|
| 105 |
-
cropped, (new_w, new_h), interpolation=cv2.INTER_AREA
|
| 106 |
-
)
|
| 107 |
-
|
| 108 |
-
pad_v = (28 - new_h) // 2
|
| 109 |
-
pad_h = (28 - new_w) // 2
|
| 110 |
-
padded = np.pad(
|
| 111 |
-
resized,
|
| 112 |
-
((pad_v, 28 - new_h - pad_v), (pad_h, 28 - new_w - pad_h)),
|
| 113 |
-
"constant",
|
| 114 |
-
)
|
| 115 |
-
else:
|
| 116 |
-
padded = cv2.resize(thresh, (28, 28))
|
| 117 |
-
|
| 118 |
-
final_img = center_image(padded)
|
| 119 |
-
input_data = (final_img.astype("float32") / 255.0).reshape(1, 28, 28, 1)
|
| 120 |
-
|
| 121 |
-
pred = model.predict(input_data, verbose=0)[0]
|
| 122 |
-
return final_img, {str(i): float(pred[i]) for i in range(10)}
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
# واجهة Gradio
|
| 126 |
-
with gr.Blocks(title="MNIST Predictor") as demo:
|
| 127 |
-
gr.Markdown("# 🧠 التنبؤ بالأرقام المرسومة (MNIST)")
|
| 128 |
|
| 129 |
with gr.Row():
|
| 130 |
-
with gr.Column():
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
)
|
| 143 |
-
|
| 144 |
-
if __name__ == "__main__":
|
| 145 |
-
demo = gr.Interface(fn=predict_digit, inputs="sketchpad", outputs="label")
|
| 146 |
-
demo.launch()
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import numpy as np
|
| 3 |
import tensorflow as tf
|
| 4 |
+
import cv2
|
| 5 |
+
from tensorflow.keras import datasets, layers, models
|
| 6 |
+
import os
|
| 7 |
+
import spaces # <--- 1. استيراد مكتبة spaces
|
| 8 |
+
|
| 9 |
+
# ============================================
|
| 10 |
+
# 1. تحميل النموذج في النطاق العام (مرة واحدة فقط)
|
| 11 |
+
# ============================================
|
| 12 |
+
model_path = 'mnist_cnn_model.keras'
|
| 13 |
+
|
| 14 |
+
if os.path.exists(model_path):
|
| 15 |
+
model = tf.keras.models.load_model(model_path)
|
| 16 |
+
print("✅ تم تحميل النموذج المحفوظ.")
|
| 17 |
+
else:
|
| 18 |
+
print("⚠️ لم يتم العثور على النموذج، جارٍ البناء والتدريب...")
|
| 19 |
+
model = models.Sequential([
|
| 20 |
+
layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
|
| 21 |
+
layers.MaxPooling2D((2, 2)),
|
| 22 |
+
layers.Conv2D(64, (3, 3), activation='relu'),
|
| 23 |
+
layers.MaxPooling2D((2, 2)),
|
| 24 |
+
layers.Flatten(),
|
| 25 |
+
layers.Dense(64, activation='relu'),
|
| 26 |
+
layers.Dense(10, activation='softmax')
|
| 27 |
+
])
|
| 28 |
+
model.compile(optimizer='adam',
|
| 29 |
+
loss='sparse_categorical_crossentropy',
|
| 30 |
+
metrics=['accuracy'])
|
| 31 |
+
|
| 32 |
+
(x_train, y_train), _ = datasets.mnist.load_data()
|
| 33 |
+
x_train = x_train.reshape((60000, 28, 28, 1)).astype('float32') / 255
|
| 34 |
+
model.fit(x_train, y_train, epochs=3, validation_split=0.1, verbose=1)
|
| 35 |
+
model.save(model_path)
|
| 36 |
+
print("✅ تم بناء النموذج وتدريبه وحفظه.")
|
| 37 |
+
|
| 38 |
+
# ============================================
|
| 39 |
+
# 2. تحميل بيانات الاختبار للأمثلة العشوائية
|
| 40 |
+
# ============================================
|
| 41 |
+
(_, _), (test_images, test_labels) = datasets.mnist.load_data()
|
| 42 |
+
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255
|
| 43 |
+
|
| 44 |
+
# ============================================
|
| 45 |
+
# 3. دوال التنبؤ والمثال العشوائي
|
| 46 |
+
# ============================================
|
| 47 |
+
# --- الدالة التي تستخدم GPU يتم تزيينها بـ @spaces.GPU ---
|
| 48 |
@spaces.GPU
|
| 49 |
+
def predict_sketch(image):
|
| 50 |
+
try:
|
| 51 |
+
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
| 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 |
+
pred = model.predict(reshaped, verbose=0)[0]
|
| 58 |
+
return {str(i): float(pred[i]) for i in range(10)}
|
| 59 |
+
except Exception as e:
|
| 60 |
+
return {"خطأ": str(e)}
|
| 61 |
+
|
| 62 |
+
# --- هذه الدالة لا تحتاج GPU، لذا لا تزينها ---
|
| 63 |
+
def random_example():
|
| 64 |
+
idx = np.random.randint(0, len(test_images))
|
| 65 |
+
img = test_images[idx].reshape(28, 28)
|
| 66 |
+
img_rgb = np.stack([img] * 3, axis=2)
|
| 67 |
+
return img_rgb, f"الرقم الحقيقي: {test_labels[idx]}"
|
| 68 |
+
|
| 69 |
+
# ============================================
|
| 70 |
+
# 4. بناء واجهة Gradio
|
| 71 |
+
# ============================================
|
| 72 |
+
with gr.Blocks(title="MNIST Recognizer") as demo:
|
| 73 |
+
gr.Markdown("# 🧠 التعرف على الأرقام المكتوبة بخط اليد")
|
| 74 |
+
gr.Markdown("ارسم رقماً (0-9) في المربع، أو اضغط على زر **مثال عشوائي**.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
|
| 76 |
with gr.Row():
|
| 77 |
+
with gr.Column(scale=1):
|
| 78 |
+
sketch = gr.Sketchpad(label="✏️ ارسم هنا")
|
| 79 |
+
with gr.Row():
|
| 80 |
+
submit_btn = gr.Button("🔮 توقع", variant="primary")
|
| 81 |
+
random_btn = gr.Button("🎲 مثال عشوائي", variant="secondary")
|
| 82 |
+
info = gr.Textbox(label="📌 معلومات", interactive=False)
|
| 83 |
+
|
| 84 |
+
with gr.Column(scale=1):
|
| 85 |
+
output = gr.Label(num_top_classes=3, label="📊 الاحتمالات")
|
| 86 |
+
|
| 87 |
+
submit_btn.click(fn=predict_sketch, inputs=sketch, outputs=output)
|
| 88 |
+
random_btn.click(fn=random_example, inputs=[], outputs=[sketch, info])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|