Update app.py
Browse files
app.py
CHANGED
|
@@ -1,38 +1,25 @@
|
|
| 1 |
import os
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
os.environ["CPP_MIN_LOG_LEVEL"] = "3"
|
| 8 |
-
|
| 9 |
-
import logging
|
| 10 |
-
import re
|
| 11 |
import gradio as gr
|
| 12 |
-
from paddleocr import PaddleOCR
|
| 13 |
from PIL import Image, ImageDraw, ImageFont
|
| 14 |
-
|
| 15 |
-
import requests
|
| 16 |
-
|
| 17 |
-
# Tắt log thừa
|
| 18 |
-
logging.getLogger("ppocr").setLevel(logging.WARNING)
|
| 19 |
-
|
| 20 |
-
print("Đang khởi tạo PaddleOCR (Coordinate Sync Mode)...")
|
| 21 |
-
|
| 22 |
-
try:
|
| 23 |
-
ocr = PaddleOCR(use_textline_orientation=True, use_doc_orientation_classify=False,
|
| 24 |
-
use_doc_unwarping=False, lang='ch')
|
| 25 |
-
except Exception as e:
|
| 26 |
-
print(f"Lỗi khởi tạo: {e}. Chuyển về chế độ mặc định.")
|
| 27 |
-
ocr = PaddleOCR(lang='ch')
|
| 28 |
|
| 29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
|
| 31 |
-
#
|
| 32 |
def check_and_download_font():
|
| 33 |
font_path = "./simfang.ttf"
|
| 34 |
if not os.path.exists(font_path):
|
| 35 |
try:
|
|
|
|
| 36 |
url = "https://github.com/StellarCN/scp_zh/raw/master/fonts/SimFang.ttf"
|
| 37 |
r = requests.get(url, allow_redirects=True)
|
| 38 |
with open(font_path, 'wb') as f:
|
|
@@ -43,179 +30,182 @@ def check_and_download_font():
|
|
| 43 |
|
| 44 |
FONT_PATH = check_and_download_font()
|
| 45 |
|
| 46 |
-
#
|
| 47 |
-
def
|
| 48 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
if isinstance(image, np.ndarray):
|
| 52 |
image = Image.fromarray(image)
|
| 53 |
-
|
| 54 |
-
# Copy để vẽ
|
| 55 |
-
canvas = image.copy()
|
| 56 |
-
draw = ImageDraw.Draw(canvas)
|
| 57 |
|
| 58 |
try:
|
| 59 |
-
|
| 60 |
-
font = ImageFont.truetype(font_path, font_size) if font_path else ImageFont.load_default()
|
| 61 |
except:
|
| 62 |
font = ImageFont.load_default()
|
| 63 |
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
if isinstance(raw_data, list) and len(raw_data) > 0 and isinstance(raw_data[0], dict):
|
| 80 |
-
data_dict = raw_data[0]
|
| 81 |
-
texts = data_dict.get('rec_texts')
|
| 82 |
-
boxes = data_dict.get('dt_polys', data_dict.get('rec_polys', data_dict.get('dt_boxes')))
|
| 83 |
-
|
| 84 |
-
if texts and boxes and isinstance(texts, list) and isinstance(boxes, list):
|
| 85 |
-
for i in range(min(len(texts), len(boxes))):
|
| 86 |
-
txt = texts[i]
|
| 87 |
-
box = parse_box(boxes[i])
|
| 88 |
-
if box and txt: items_to_draw.append((box, txt))
|
| 89 |
-
processed = True
|
| 90 |
-
|
| 91 |
-
# Fallback Logic
|
| 92 |
-
if not processed:
|
| 93 |
-
def hunt(data):
|
| 94 |
-
if isinstance(data, dict):
|
| 95 |
-
box = None; text = None
|
| 96 |
-
for k in ['points', 'box', 'dt_boxes', 'poly']:
|
| 97 |
-
if k in data: box = parse_box(data[k]); break
|
| 98 |
-
for k in ['transcription', 'text', 'rec_text', 'label']:
|
| 99 |
-
if k in data: text = data[k]; break
|
| 100 |
-
if box and text: items_to_draw.append((box, text)); return
|
| 101 |
-
for v in data.values(): hunt(v)
|
| 102 |
-
elif isinstance(data, (list, tuple)):
|
| 103 |
-
if len(data) == 2 and isinstance(data[0], list) and len(data[0]) == 4:
|
| 104 |
-
box = parse_box(data[0])
|
| 105 |
-
txt_obj = data[1]
|
| 106 |
-
text = txt_obj[0] if isinstance(txt_obj, (list, tuple)) else txt_obj
|
| 107 |
-
if box and isinstance(text, str): items_to_draw.append((box, text)); return
|
| 108 |
-
for item in data: hunt(item)
|
| 109 |
-
hunt(raw_data)
|
| 110 |
-
|
| 111 |
-
# Vẽ
|
| 112 |
-
for box, txt in items_to_draw:
|
| 113 |
-
try:
|
| 114 |
-
# Vẽ khung đỏ
|
| 115 |
-
draw.polygon(box, outline="red", width=3)
|
| 116 |
# Vẽ chữ
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
for val in data.values(): found_texts.extend(deep_extract_text(val))
|
| 138 |
-
elif hasattr(data, '__dict__'): found_texts.extend(deep_extract_text(data.__dict__))
|
| 139 |
-
return found_texts
|
| 140 |
-
|
| 141 |
-
def clean_text_result(text_list):
|
| 142 |
-
cleaned = []
|
| 143 |
-
block_list = ['min', 'max', 'general', 'header', 'footer', 'structure']
|
| 144 |
-
for t in text_list:
|
| 145 |
-
t = t.strip()
|
| 146 |
-
if len(t) < 2 and not any(u'\u4e00' <= c <= u'\u9fff' for c in t): continue
|
| 147 |
-
if t.lower().endswith(('.ttf', '.json', '.pdparams', '.yml', '.log')): continue
|
| 148 |
-
if t.lower() in block_list: continue
|
| 149 |
-
if not re.search(r'[\w\u4e00-\u9fff]', t): continue
|
| 150 |
-
cleaned.append(t)
|
| 151 |
-
return cleaned
|
| 152 |
-
|
| 153 |
-
# --- MAIN PREDICT ---
|
| 154 |
-
def predict(image):
|
| 155 |
-
if image is None: return None, "Chưa có ảnh.", "No Data"
|
| 156 |
-
|
| 157 |
-
try:
|
| 158 |
-
# Chuẩn bị ảnh đầu vào
|
| 159 |
-
original_pil = image.copy() if isinstance(image, Image.Image) else Image.fromarray(image).copy()
|
| 160 |
-
image_np = np.array(image)
|
| 161 |
|
| 162 |
-
|
| 163 |
-
raw_result = ocr.ocr(image_np)
|
| 164 |
|
| 165 |
-
|
| 166 |
-
|
|
|
|
|
|
|
|
|
|
| 167 |
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
debug_info = f"Used Image Source: {'Preprocessed' if target_image_for_drawing != original_pil else 'Original'}\nData Preview:\n{debug_str}..."
|
| 189 |
-
|
| 190 |
-
return annotated_image, text_output, debug_info
|
| 191 |
-
|
| 192 |
-
except Exception as e:
|
| 193 |
-
import traceback
|
| 194 |
-
return image, f"Lỗi: {str(e)}", traceback.format_exc()
|
| 195 |
-
|
| 196 |
-
# --- GIAO DIỆN ---
|
| 197 |
-
with gr.Blocks(title="PaddleOCR Perfect Overlay") as iface:
|
| 198 |
-
gr.Markdown("## PaddleOCR Chinese - High Precision Overlay")
|
| 199 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 200 |
with gr.Row():
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
with gr.Tabs():
|
| 207 |
-
with gr.TabItem("🖼️
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
|
|
|
|
|
|
|
|
|
| 214 |
submit_btn.click(
|
| 215 |
-
fn=
|
| 216 |
-
inputs=
|
| 217 |
-
outputs=[
|
| 218 |
)
|
| 219 |
|
| 220 |
if __name__ == "__main__":
|
| 221 |
-
|
|
|
|
| 1 |
import os
|
| 2 |
+
import cv2
|
| 3 |
+
import json
|
| 4 |
+
import tarfile
|
| 5 |
+
import requests
|
| 6 |
+
import numpy as np
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
import gradio as gr
|
|
|
|
| 8 |
from PIL import Image, ImageDraw, ImageFont
|
| 9 |
+
from paddleocr import PaddleOCR
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
+
# ==========================================
|
| 12 |
+
# 1. CẤU HÌNH & TẢI MODEL (SERVER VERSION)
|
| 13 |
+
# ==========================================
|
| 14 |
+
os.environ["FLAGS_use_mkldnn"] = "0"
|
| 15 |
+
os.environ["CPP_MIN_LOG_LEVEL"] = "3"
|
| 16 |
|
| 17 |
+
# Hàm tải font chữ Trung Quốc (Giữ lại từ code cũ của bạn vì rất tốt)
|
| 18 |
def check_and_download_font():
|
| 19 |
font_path = "./simfang.ttf"
|
| 20 |
if not os.path.exists(font_path):
|
| 21 |
try:
|
| 22 |
+
print("Đang tải font SimFang...")
|
| 23 |
url = "https://github.com/StellarCN/scp_zh/raw/master/fonts/SimFang.ttf"
|
| 24 |
r = requests.get(url, allow_redirects=True)
|
| 25 |
with open(font_path, 'wb') as f:
|
|
|
|
| 30 |
|
| 31 |
FONT_PATH = check_and_download_font()
|
| 32 |
|
| 33 |
+
# Hàm tải Model Server (Độ chính xác cao)
|
| 34 |
+
def download_model_server(save_dir="./server_models"):
|
| 35 |
+
urls = {
|
| 36 |
+
"det": "https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_det_server_infer.tar",
|
| 37 |
+
"rec": "https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_server_infer.tar",
|
| 38 |
+
"cls": "https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar"
|
| 39 |
+
}
|
| 40 |
|
| 41 |
+
paths = {}
|
| 42 |
+
if not os.path.exists(save_dir): os.makedirs(save_dir)
|
| 43 |
+
|
| 44 |
+
for key, url in urls.items():
|
| 45 |
+
filename = url.split("/")[-1]
|
| 46 |
+
extract_name = filename.replace('.tar', '')
|
| 47 |
+
full_path = os.path.join(save_dir, extract_name)
|
| 48 |
+
|
| 49 |
+
if not os.path.exists(full_path):
|
| 50 |
+
print(f"Đang tải model {key.upper()} High-Accuracy...")
|
| 51 |
+
tar_path = os.path.join(save_dir, filename)
|
| 52 |
+
r = requests.get(url, stream=True)
|
| 53 |
+
with open(tar_path, 'wb') as f:
|
| 54 |
+
for chunk in r.iter_content(chunk_size=1024):
|
| 55 |
+
if chunk: f.write(chunk)
|
| 56 |
+
with tarfile.open(tar_path) as tar:
|
| 57 |
+
tar.extractall(path=save_dir)
|
| 58 |
+
os.remove(tar_path)
|
| 59 |
+
paths[key] = full_path
|
| 60 |
+
return paths
|
| 61 |
+
|
| 62 |
+
# Khởi tạo OCR
|
| 63 |
+
print("Đang khởi tạo PaddleOCR Server Mode...")
|
| 64 |
+
try:
|
| 65 |
+
models = download_model_server()
|
| 66 |
+
ocr = PaddleOCR(
|
| 67 |
+
use_angle_cls=True, lang='ch',
|
| 68 |
+
det_model_dir=models['det'],
|
| 69 |
+
rec_model_dir=models['rec'],
|
| 70 |
+
cls_model_dir=models['cls'],
|
| 71 |
+
use_textline_orientation=True
|
| 72 |
+
)
|
| 73 |
+
print("Model Server đã sẵn sàng!")
|
| 74 |
+
except Exception as e:
|
| 75 |
+
print(f"Lỗi tải model server: {e}. Dùng Mobile model.")
|
| 76 |
+
ocr = PaddleOCR(use_angle_cls=True, lang='ch')
|
| 77 |
+
|
| 78 |
+
# ==========================================
|
| 79 |
+
# 2. XỬ LÝ HÌNH ẢNH & KẾT QUẢ
|
| 80 |
+
# ==========================================
|
| 81 |
+
|
| 82 |
+
def draw_results(image, result, font_path):
|
| 83 |
+
# Convert sang PIL để vẽ đẹp hơn
|
| 84 |
if isinstance(image, np.ndarray):
|
| 85 |
image = Image.fromarray(image)
|
| 86 |
+
draw = ImageDraw.Draw(image)
|
|
|
|
|
|
|
|
|
|
| 87 |
|
| 88 |
try:
|
| 89 |
+
font = ImageFont.truetype(font_path, 20) if font_path else ImageFont.load_default()
|
|
|
|
| 90 |
except:
|
| 91 |
font = ImageFont.load_default()
|
| 92 |
|
| 93 |
+
if result and result[0]:
|
| 94 |
+
for line in result[0]:
|
| 95 |
+
box = np.array(line[0]).astype(np.int32)
|
| 96 |
+
txt = line[1][0]
|
| 97 |
+
conf = line[1][1]
|
| 98 |
+
|
| 99 |
+
# Vẽ box
|
| 100 |
+
tuples = [tuple(p) for p in box]
|
| 101 |
+
draw.polygon(tuples, outline="red", width=2)
|
| 102 |
+
|
| 103 |
+
# Vẽ nền chữ
|
| 104 |
+
txt_pos = (box[0][0], box[0][1] - 25)
|
| 105 |
+
bbox = draw.textbbox(txt_pos, f"{txt} ({conf:.2f})", font=font)
|
| 106 |
+
draw.rectangle(bbox, fill="red")
|
| 107 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
# Vẽ chữ
|
| 109 |
+
draw.text(txt_pos, txt, fill="white", font=font)
|
| 110 |
+
|
| 111 |
+
return image
|
| 112 |
+
|
| 113 |
+
def format_output(result):
|
| 114 |
+
"""Chuyển đổi kết quả sang Markdown và JSON sạch"""
|
| 115 |
+
if not result or not result[0]:
|
| 116 |
+
return "Không tìm thấy văn bản.", "[]"
|
| 117 |
+
|
| 118 |
+
# Tạo Markdown
|
| 119 |
+
md_lines = []
|
| 120 |
+
json_data = []
|
| 121 |
+
|
| 122 |
+
# Sắp xếp theo trục Y để tạo dòng văn bản tự nhiên
|
| 123 |
+
boxes = sorted(result[0], key=lambda x: x[0][0][1])
|
| 124 |
+
|
| 125 |
+
for item in boxes:
|
| 126 |
+
text = item[1][0]
|
| 127 |
+
conf = float(item[1][1])
|
| 128 |
+
box = item[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
md_lines.append(f"- **{text}** (Độ tin cậy: {conf:.1%})")
|
|
|
|
| 131 |
|
| 132 |
+
json_data.append({
|
| 133 |
+
"text": text,
|
| 134 |
+
"confidence": conf,
|
| 135 |
+
"box": box
|
| 136 |
+
})
|
| 137 |
|
| 138 |
+
md_output = "\n".join(md_lines)
|
| 139 |
+
json_output = json.dumps(json_data, ensure_ascii=False, indent=2)
|
| 140 |
+
|
| 141 |
+
return md_output, json_output
|
| 142 |
+
|
| 143 |
+
# ==========================================
|
| 144 |
+
# 3. HÀM DỰ ĐOÁN CHÍNH
|
| 145 |
+
# ==========================================
|
| 146 |
+
def predict_pipeline(image_file):
|
| 147 |
+
if image_file is None:
|
| 148 |
+
return None, "", ""
|
| 149 |
+
|
| 150 |
+
# Đọc ảnh
|
| 151 |
+
img = np.array(Image.open(image_file).convert('RGB'))
|
| 152 |
+
|
| 153 |
+
# OCR
|
| 154 |
+
result = ocr.ocr(img, cls=True)
|
| 155 |
+
|
| 156 |
+
# 1. Vẽ Visualization
|
| 157 |
+
vis_img = draw_results(img.copy(), result, FONT_PATH)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 158 |
|
| 159 |
+
# 2. Format dữ liệu
|
| 160 |
+
md_out, json_out = format_output(result)
|
| 161 |
+
|
| 162 |
+
return vis_img, md_out, json_out
|
| 163 |
+
|
| 164 |
+
# ==========================================
|
| 165 |
+
# 4. GIAO DIỆN GRADIO (Custom CSS giống bản Demo)
|
| 166 |
+
# ==========================================
|
| 167 |
+
custom_css = """
|
| 168 |
+
body, .gradio-container { font-family: "Noto Sans SC", sans-serif; }
|
| 169 |
+
.gradio-container { max-width: 1200px !important; margin: auto; }
|
| 170 |
+
.header-area { text-align: center; margin-bottom: 20px; }
|
| 171 |
+
.header-area h1 { margin-bottom: 5px; color: #2d3748; }
|
| 172 |
+
.notice { background: #f0f9ff; border: 1px solid #bae6fd; padding: 10px; border-radius: 8px; color: #0369a1; font-size: 14px; margin-bottom: 15px; }
|
| 173 |
+
"""
|
| 174 |
+
|
| 175 |
+
with gr.Blocks(title="PaddleOCR Pro Local", css=custom_css, theme=gr.themes.Soft()) as app:
|
| 176 |
+
|
| 177 |
+
with gr.Column(elem_classes="header-area"):
|
| 178 |
+
gr.Markdown("# 🇨🇳 PaddleOCR Professional (Local Version)")
|
| 179 |
+
gr.HTML("<div class='notice'>⚡ Phiên bản Server-Mode: Chạy offline với độ chính xác cao hơn bản Mobile mặc định.</div>")
|
| 180 |
+
|
| 181 |
with gr.Row():
|
| 182 |
+
# Cột TRÁI: Input
|
| 183 |
+
with gr.Column(scale=4):
|
| 184 |
+
input_image = gr.Image(type="filepath", label="Tải ảnh lên", height=400)
|
| 185 |
+
submit_btn = gr.Button("🚀 CHẠY NHẬN DIỆN", variant="primary", size="lg")
|
| 186 |
+
|
| 187 |
+
gr.Markdown("### 💡 Ghi chú:")
|
| 188 |
+
gr.Markdown("- Model sẽ tự động tải phiên bản **Server (High Accuracy)** (~200MB) trong lần chạy đầu.")
|
| 189 |
+
gr.Markdown("- Hỗ trợ tốt cho tài liệu scan, hóa đơn và văn bản tiếng Trung.")
|
| 190 |
+
|
| 191 |
+
# Cột PHẢI: Output (Tabbed UI)
|
| 192 |
+
with gr.Column(scale=6):
|
| 193 |
with gr.Tabs():
|
| 194 |
+
with gr.TabItem("🖼️ Trực quan hóa (Visualization)"):
|
| 195 |
+
output_image = gr.Image(type="pil", label="Kết quả")
|
| 196 |
+
|
| 197 |
+
with gr.TabItem("📝 Văn bản (Markdown)"):
|
| 198 |
+
output_md = gr.Markdown(label="Nội dung trích xuất")
|
| 199 |
+
|
| 200 |
+
with gr.TabItem("📊 Dữ liệu thô (JSON)"):
|
| 201 |
+
output_json = gr.Code(language="json", label="Chi tiết tọa độ & Confidence")
|
| 202 |
+
|
| 203 |
+
# Xử lý sự kiện
|
| 204 |
submit_btn.click(
|
| 205 |
+
fn=predict_pipeline,
|
| 206 |
+
inputs=[input_image],
|
| 207 |
+
outputs=[output_image, output_md, output_json]
|
| 208 |
)
|
| 209 |
|
| 210 |
if __name__ == "__main__":
|
| 211 |
+
app.launch(server_name="0.0.0.0", server_port=7860)
|