Update app.py
Browse files
app.py
CHANGED
|
@@ -4,6 +4,7 @@ import os
|
|
| 4 |
import cv2
|
| 5 |
import numpy as np
|
| 6 |
import gradio as gr
|
|
|
|
| 7 |
from ultralytics import YOLO
|
| 8 |
from PIL import Image, ImageDraw, ImageFont
|
| 9 |
import arabic_reshaper
|
|
@@ -13,19 +14,29 @@ import google.generativeai as genai
|
|
| 13 |
import torch
|
| 14 |
|
| 15 |
# ==========================
|
| 16 |
-
# ⚠️
|
| 17 |
# ==========================
|
| 18 |
|
| 19 |
-
GEMINI_API_KEY = "YOUR_GEMINI_API_KEY_HERE" #
|
| 20 |
-
|
| 21 |
genai.configure(api_key=GEMINI_API_KEY)
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
# ==========================
|
| 24 |
# إعدادات أداء PyTorch / GPU
|
| 25 |
# ==========================
|
| 26 |
|
| 27 |
-
#
|
| 28 |
-
torch.backends.cudnn.benchmark = True
|
| 29 |
|
| 30 |
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
|
| 31 |
USE_HALF = DEVICE.startswith("cuda")
|
|
@@ -53,10 +64,6 @@ arabic_map = {
|
|
| 53 |
"toot": "ة", "al": "ال"
|
| 54 |
}
|
| 55 |
|
| 56 |
-
# ==========================
|
| 57 |
-
# نص توجيهات Gemini
|
| 58 |
-
# ==========================
|
| 59 |
-
|
| 60 |
SYSTEM_PROMPT = (
|
| 61 |
"أنت مساعد ذكي يستقبل كلمات أو جمل قصيرة قادمة من مترجم لغة "
|
| 62 |
"الإشارة العربية، ودورك أن تعيد صياغتها كنص عربي واضح ومفهوم، "
|
|
@@ -64,7 +71,7 @@ SYSTEM_PROMPT = (
|
|
| 64 |
)
|
| 65 |
|
| 66 |
# ==========================
|
| 67 |
-
# إعداد خط عربي مرة وحدة
|
| 68 |
# ==========================
|
| 69 |
|
| 70 |
DEFAULT_FONT_SIZE = 24
|
|
@@ -77,21 +84,15 @@ except Exception:
|
|
| 77 |
|
| 78 |
|
| 79 |
def prepare_arabic(text: str) -> str:
|
| 80 |
-
"""تجهيز النص العربي (reshaper + bidi) بدون رسم."""
|
| 81 |
reshaped = arabic_reshaper.reshape(text)
|
| 82 |
bidi_text = get_display(reshaped)
|
| 83 |
return bidi_text
|
| 84 |
|
| 85 |
-
|
| 86 |
# ==========================
|
| 87 |
-
# رسم الديتكشن بشكل
|
| 88 |
# ==========================
|
| 89 |
|
| 90 |
def draw_detections(result, frame, names):
|
| 91 |
-
"""
|
| 92 |
-
نرسم البوكسات بـ OpenCV،
|
| 93 |
-
وبعدين نحول الصورة إلى PIL مرة واحدة ونرسم كل النصوص.
|
| 94 |
-
"""
|
| 95 |
boxes = result.boxes
|
| 96 |
detected_labels = []
|
| 97 |
|
|
@@ -112,7 +113,7 @@ def draw_detections(result, frame, names):
|
|
| 112 |
ar_label = arabic_map.get(eng_label, eng_label)
|
| 113 |
detected_labels.append(ar_label)
|
| 114 |
|
| 115 |
-
#
|
| 116 |
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
|
| 117 |
|
| 118 |
label_bg_y1 = max(0, y1 - 35)
|
|
@@ -127,13 +128,12 @@ def draw_detections(result, frame, names):
|
|
| 127 |
|
| 128 |
label_infos.append(
|
| 129 |
(
|
| 130 |
-
prepare_arabic(ar_label),
|
| 131 |
x1 + 5,
|
| 132 |
label_bg_y1 + 5,
|
| 133 |
)
|
| 134 |
)
|
| 135 |
|
| 136 |
-
# نحول لمرة وحدة إلى PIL ونرسم كل النصوص
|
| 137 |
img_pil = Image.fromarray(frame)
|
| 138 |
draw = ImageDraw.Draw(img_pil)
|
| 139 |
|
|
@@ -142,7 +142,6 @@ def draw_detections(result, frame, names):
|
|
| 142 |
|
| 143 |
return np.array(img_pil), detected_labels
|
| 144 |
|
| 145 |
-
|
| 146 |
# ==========================
|
| 147 |
# تحميل YOLO على GPU + half
|
| 148 |
# ==========================
|
|
@@ -151,7 +150,6 @@ print("🔹 Loading YOLO model...")
|
|
| 151 |
model = YOLO(WEIGHTS_PATH)
|
| 152 |
model.to(DEVICE)
|
| 153 |
|
| 154 |
-
# تفعيل half precision لو معنا GPU
|
| 155 |
if USE_HALF:
|
| 156 |
try:
|
| 157 |
model.model.half()
|
|
@@ -161,7 +159,6 @@ if USE_HALF:
|
|
| 161 |
|
| 162 |
print("📚 Classes:", model.names)
|
| 163 |
|
| 164 |
-
|
| 165 |
# ==========================
|
| 166 |
# Gemini API Call
|
| 167 |
# ==========================
|
|
@@ -171,7 +168,6 @@ def call_gemini_on_word(word: str) -> str:
|
|
| 171 |
return ""
|
| 172 |
|
| 173 |
try:
|
| 174 |
-
# flash سريع أصلا، نخليه كما هو
|
| 175 |
model_g = genai.GenerativeModel("gemini-1.5-flash")
|
| 176 |
|
| 177 |
prompt = (
|
|
@@ -185,7 +181,6 @@ def call_gemini_on_word(word: str) -> str:
|
|
| 185 |
except Exception as e:
|
| 186 |
return f"خطأ Gemini: {e}"
|
| 187 |
|
| 188 |
-
|
| 189 |
# ==========================
|
| 190 |
# معالجة الفريم
|
| 191 |
# ==========================
|
|
@@ -201,12 +196,9 @@ def process_frame(
|
|
| 201 |
if chat_history is None:
|
| 202 |
chat_history = []
|
| 203 |
|
| 204 |
-
# Gradio يعطينا RGB، نحوله BGR لـ OpenCV
|
| 205 |
frame_bgr = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
|
| 206 |
-
# نعكس الصورة أفقياً
|
| 207 |
frame_bgr = cv2.flip(frame_bgr, 1)
|
| 208 |
|
| 209 |
-
# YOLO على GPU + نصف دقة لو متوفر
|
| 210 |
results = model.predict(
|
| 211 |
frame_bgr,
|
| 212 |
conf=CONF_THRESHOLD,
|
|
@@ -269,7 +261,6 @@ def process_frame(
|
|
| 269 |
chat_history,
|
| 270 |
)
|
| 271 |
|
| 272 |
-
|
| 273 |
# ==========================
|
| 274 |
# واجه�� Gradio
|
| 275 |
# ==========================
|
|
@@ -317,8 +308,6 @@ with gr.Blocks() as demo:
|
|
| 317 |
],
|
| 318 |
)
|
| 319 |
|
| 320 |
-
|
| 321 |
if __name__ == "__main__":
|
| 322 |
-
|
| 323 |
-
# نقدر نضيف باتش بسيط هنا، بس حالياً نخليها عادية:
|
| 324 |
-
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 4 |
import cv2
|
| 5 |
import numpy as np
|
| 6 |
import gradio as gr
|
| 7 |
+
import gradio.utils as gr_utils
|
| 8 |
from ultralytics import YOLO
|
| 9 |
from PIL import Image, ImageDraw, ImageFont
|
| 10 |
import arabic_reshaper
|
|
|
|
| 14 |
import torch
|
| 15 |
|
| 16 |
# ==========================
|
| 17 |
+
# ⚠️ Gemini API Key
|
| 18 |
# ==========================
|
| 19 |
|
| 20 |
+
GEMINI_API_KEY = "YOUR_GEMINI_API_KEY_HERE" # ← حط المفتاح هنا
|
|
|
|
| 21 |
genai.configure(api_key=GEMINI_API_KEY)
|
| 22 |
|
| 23 |
+
# ==========================
|
| 24 |
+
# Patch لمشكلة Spaces hot-reload
|
| 25 |
+
# ==========================
|
| 26 |
+
|
| 27 |
+
# داخل Hugging Face Spaces، SPACE_ID يكون موجود في الـ env
|
| 28 |
+
if os.getenv("SPACE_ID"):
|
| 29 |
+
def _no_watchfn_spaces(*args, **kwargs):
|
| 30 |
+
# نطفي الـ hot-reload اللي يسبب RuntimeError
|
| 31 |
+
return
|
| 32 |
+
# نكتب فوق الدالة الأصلية
|
| 33 |
+
gr_utils.watchfn_spaces = _no_watchfn_spaces
|
| 34 |
+
|
| 35 |
# ==========================
|
| 36 |
# إعدادات أداء PyTorch / GPU
|
| 37 |
# ==========================
|
| 38 |
|
| 39 |
+
torch.backends.cudnn.benchmark = True # تسريع الـ conv على GPU
|
|
|
|
| 40 |
|
| 41 |
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
|
| 42 |
USE_HALF = DEVICE.startswith("cuda")
|
|
|
|
| 64 |
"toot": "ة", "al": "ال"
|
| 65 |
}
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
SYSTEM_PROMPT = (
|
| 68 |
"أنت مساعد ذكي يستقبل كلمات أو جمل قصيرة قادمة من مترجم لغة "
|
| 69 |
"الإشارة العربية، ودورك أن تعيد صياغتها كنص عربي واضح ومفهوم، "
|
|
|
|
| 71 |
)
|
| 72 |
|
| 73 |
# ==========================
|
| 74 |
+
# إعداد خط عربي مرة وحدة
|
| 75 |
# ==========================
|
| 76 |
|
| 77 |
DEFAULT_FONT_SIZE = 24
|
|
|
|
| 84 |
|
| 85 |
|
| 86 |
def prepare_arabic(text: str) -> str:
|
|
|
|
| 87 |
reshaped = arabic_reshaper.reshape(text)
|
| 88 |
bidi_text = get_display(reshaped)
|
| 89 |
return bidi_text
|
| 90 |
|
|
|
|
| 91 |
# ==========================
|
| 92 |
+
# رسم الديتكشن بشكل سريع
|
| 93 |
# ==========================
|
| 94 |
|
| 95 |
def draw_detections(result, frame, names):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
boxes = result.boxes
|
| 97 |
detected_labels = []
|
| 98 |
|
|
|
|
| 113 |
ar_label = arabic_map.get(eng_label, eng_label)
|
| 114 |
detected_labels.append(ar_label)
|
| 115 |
|
| 116 |
+
# box
|
| 117 |
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
|
| 118 |
|
| 119 |
label_bg_y1 = max(0, y1 - 35)
|
|
|
|
| 128 |
|
| 129 |
label_infos.append(
|
| 130 |
(
|
| 131 |
+
prepare_arabic(ar_label),
|
| 132 |
x1 + 5,
|
| 133 |
label_bg_y1 + 5,
|
| 134 |
)
|
| 135 |
)
|
| 136 |
|
|
|
|
| 137 |
img_pil = Image.fromarray(frame)
|
| 138 |
draw = ImageDraw.Draw(img_pil)
|
| 139 |
|
|
|
|
| 142 |
|
| 143 |
return np.array(img_pil), detected_labels
|
| 144 |
|
|
|
|
| 145 |
# ==========================
|
| 146 |
# تحميل YOLO على GPU + half
|
| 147 |
# ==========================
|
|
|
|
| 150 |
model = YOLO(WEIGHTS_PATH)
|
| 151 |
model.to(DEVICE)
|
| 152 |
|
|
|
|
| 153 |
if USE_HALF:
|
| 154 |
try:
|
| 155 |
model.model.half()
|
|
|
|
| 159 |
|
| 160 |
print("📚 Classes:", model.names)
|
| 161 |
|
|
|
|
| 162 |
# ==========================
|
| 163 |
# Gemini API Call
|
| 164 |
# ==========================
|
|
|
|
| 168 |
return ""
|
| 169 |
|
| 170 |
try:
|
|
|
|
| 171 |
model_g = genai.GenerativeModel("gemini-1.5-flash")
|
| 172 |
|
| 173 |
prompt = (
|
|
|
|
| 181 |
except Exception as e:
|
| 182 |
return f"خطأ Gemini: {e}"
|
| 183 |
|
|
|
|
| 184 |
# ==========================
|
| 185 |
# معالجة الفريم
|
| 186 |
# ==========================
|
|
|
|
| 196 |
if chat_history is None:
|
| 197 |
chat_history = []
|
| 198 |
|
|
|
|
| 199 |
frame_bgr = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
|
|
|
|
| 200 |
frame_bgr = cv2.flip(frame_bgr, 1)
|
| 201 |
|
|
|
|
| 202 |
results = model.predict(
|
| 203 |
frame_bgr,
|
| 204 |
conf=CONF_THRESHOLD,
|
|
|
|
| 261 |
chat_history,
|
| 262 |
)
|
| 263 |
|
|
|
|
| 264 |
# ==========================
|
| 265 |
# واجه�� Gradio
|
| 266 |
# ==========================
|
|
|
|
| 308 |
],
|
| 309 |
)
|
| 310 |
|
| 311 |
+
# في Spaces ما نحتاج نحدد port غالباً
|
| 312 |
if __name__ == "__main__":
|
| 313 |
+
demo.launch()
|
|
|
|
|
|