Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -7,17 +7,6 @@ import numpy as np
|
|
| 7 |
import gradio as gr
|
| 8 |
from paddleocr import PaddleOCR
|
| 9 |
|
| 10 |
-
from PIL import Image
|
| 11 |
-
|
| 12 |
-
def is_valid_image(path):
|
| 13 |
-
try:
|
| 14 |
-
img = Image.open(path)
|
| 15 |
-
img.verify()
|
| 16 |
-
return True
|
| 17 |
-
except:
|
| 18 |
-
return False
|
| 19 |
-
|
| 20 |
-
|
| 21 |
ocr = PaddleOCR(use_angle_cls=True, lang='en', det_model_dir='models/det', rec_model_dir='models/rec', cls_model_dir='models/cls')
|
| 22 |
|
| 23 |
def classify_background_color(avg_color, white_thresh=230, black_thresh=50, yellow_thresh=100):
|
|
@@ -50,48 +39,20 @@ def sample_border_color(image, box, padding=2):
|
|
| 50 |
median_color = np.median(border_pixels, axis=0)
|
| 51 |
return tuple(map(int, median_color))
|
| 52 |
|
| 53 |
-
def detect_text_boxes(image
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
if image is None or not hasattr(image, 'shape'):
|
| 57 |
-
print("Invalid image. Skipping...")
|
| 58 |
-
return []
|
| 59 |
-
|
| 60 |
-
# Resize large images to reduce memory load
|
| 61 |
-
height, width = image.shape[:2]
|
| 62 |
-
if max(height, width) > max_dim:
|
| 63 |
-
scale = max_dim / float(max(height, width))
|
| 64 |
-
image = cv2.resize(image, (int(width * scale), int(height * scale)))
|
| 65 |
-
|
| 66 |
-
# Ensure image is in RGB
|
| 67 |
-
if image.shape[2] == 1:
|
| 68 |
-
image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
|
| 69 |
-
elif image.shape[2] == 3:
|
| 70 |
-
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
| 71 |
-
|
| 72 |
-
# Call PaddleOCR correctly
|
| 73 |
-
results = ocr.ocr(image, cls=True)
|
| 74 |
-
|
| 75 |
-
if results is None or not results[0]:
|
| 76 |
-
print("No OCR results found or OCR returned None.")
|
| 77 |
-
return []
|
| 78 |
-
|
| 79 |
-
boxes = []
|
| 80 |
-
for line in results[0]:
|
| 81 |
-
box, (text, confidence) = line
|
| 82 |
-
if text.strip():
|
| 83 |
-
x_min = int(min(pt[0] for pt in box))
|
| 84 |
-
x_max = int(max(pt[0] for pt in box))
|
| 85 |
-
y_min = int(min(pt[1] for pt in box))
|
| 86 |
-
y_max = int(max(pt[1] for pt in box))
|
| 87 |
-
boxes.append(((x_min, y_min, x_max, y_max), text, confidence))
|
| 88 |
-
return boxes
|
| 89 |
-
|
| 90 |
-
except Exception as e:
|
| 91 |
-
print(f"OCR failed on image: {e}")
|
| 92 |
return []
|
| 93 |
-
|
| 94 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
def remove_text_dynamic_fill(img_path, output_path):
|
| 97 |
image = cv2.imread(img_path)
|
|
@@ -138,36 +99,16 @@ def remove_text_dynamic_fill(img_path, output_path):
|
|
| 138 |
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
|
| 139 |
cv2.imwrite(output_path, image)
|
| 140 |
|
| 141 |
-
import uuid
|
| 142 |
-
|
| 143 |
def process_folder(input_files):
|
| 144 |
temp_output = tempfile.mkdtemp()
|
| 145 |
-
|
| 146 |
for file in input_files:
|
| 147 |
filename = os.path.basename(file.name)
|
| 148 |
output_path = os.path.join(temp_output, filename)
|
| 149 |
remove_text_dynamic_fill(file.name, output_path)
|
| 150 |
|
| 151 |
-
|
| 152 |
-
zip_path = os.path.join("/tmp", f"cleaned_output_{unique_name}.zip")
|
| 153 |
-
shutil.make_archive(zip_path.replace(".zip", ""), 'zip', temp_output)
|
| 154 |
-
|
| 155 |
-
delayed_cleanup(zip_path)
|
| 156 |
return zip_path
|
| 157 |
|
| 158 |
-
|
| 159 |
-
import threading
|
| 160 |
-
import time
|
| 161 |
-
|
| 162 |
-
def delayed_cleanup(path, delay=30):
|
| 163 |
-
def cleanup():
|
| 164 |
-
time.sleep(delay)
|
| 165 |
-
if os.path.exists(path):
|
| 166 |
-
os.remove(path)
|
| 167 |
-
threading.Thread(target=cleanup).start()
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
demo = gr.Interface(
|
| 172 |
fn=process_folder,
|
| 173 |
inputs=gr.File(file_types=[".jpg", ".jpeg", ".png"], file_count="multiple", label="Upload Comic Images"),
|
|
@@ -176,4 +117,4 @@ demo = gr.Interface(
|
|
| 176 |
description="Upload comic images and get a zip of cleaned versions (text removed). Uses PaddleOCR for detection."
|
| 177 |
)
|
| 178 |
|
| 179 |
-
demo.launch()
|
|
|
|
| 7 |
import gradio as gr
|
| 8 |
from paddleocr import PaddleOCR
|
| 9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
ocr = PaddleOCR(use_angle_cls=True, lang='en', det_model_dir='models/det', rec_model_dir='models/rec', cls_model_dir='models/cls')
|
| 11 |
|
| 12 |
def classify_background_color(avg_color, white_thresh=230, black_thresh=50, yellow_thresh=100):
|
|
|
|
| 39 |
median_color = np.median(border_pixels, axis=0)
|
| 40 |
return tuple(map(int, median_color))
|
| 41 |
|
| 42 |
+
def detect_text_boxes(image):
|
| 43 |
+
results = ocr.ocr(image, cls=True)
|
| 44 |
+
if not results or not results[0]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
return []
|
| 46 |
+
boxes = []
|
| 47 |
+
for line in results[0]:
|
| 48 |
+
box, (text, confidence) = line
|
| 49 |
+
if text.strip():
|
| 50 |
+
x_min = int(min(pt[0] for pt in box))
|
| 51 |
+
x_max = int(max(pt[0] for pt in box))
|
| 52 |
+
y_min = int(min(pt[1] for pt in box))
|
| 53 |
+
y_max = int(max(pt[1] for pt in box))
|
| 54 |
+
boxes.append(((x_min, y_min, x_max, y_max), text, confidence))
|
| 55 |
+
return boxes
|
| 56 |
|
| 57 |
def remove_text_dynamic_fill(img_path, output_path):
|
| 58 |
image = cv2.imread(img_path)
|
|
|
|
| 99 |
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
|
| 100 |
cv2.imwrite(output_path, image)
|
| 101 |
|
|
|
|
|
|
|
| 102 |
def process_folder(input_files):
|
| 103 |
temp_output = tempfile.mkdtemp()
|
|
|
|
| 104 |
for file in input_files:
|
| 105 |
filename = os.path.basename(file.name)
|
| 106 |
output_path = os.path.join(temp_output, filename)
|
| 107 |
remove_text_dynamic_fill(file.name, output_path)
|
| 108 |
|
| 109 |
+
zip_path = shutil.make_archive(temp_output, 'zip', temp_output)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
return zip_path
|
| 111 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
demo = gr.Interface(
|
| 113 |
fn=process_folder,
|
| 114 |
inputs=gr.File(file_types=[".jpg", ".jpeg", ".png"], file_count="multiple", label="Upload Comic Images"),
|
|
|
|
| 117 |
description="Upload comic images and get a zip of cleaned versions (text removed). Uses PaddleOCR for detection."
|
| 118 |
)
|
| 119 |
|
| 120 |
+
demo.launch()
|