Spaces:
Sleeping
Sleeping
File size: 13,612 Bytes
8804c8f a9ccd5e 8804c8f a9ccd5e 8804c8f a9ccd5e 8804c8f a9ccd5e 8804c8f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 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 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 |
from recognize import recongize
from ner import ner
import os
import time
import argparse
from sr.sr import sr
import torch
from scipy.ndimage import gaussian_filter
from PIL import Image
import numpy as np
import torch.nn as nn
import torch.backends.cudnn as cudnn
from torch.autograd import Variable
from mosaik import mosaik
from PIL import Image
import cv2
from skimage import io
import numpy as np
import craft_utils
import imgproc
import file_utils
from seg import mask_percentage
from seg2 import dino_seg
from craft import CRAFT
from collections import OrderedDict
import gradio as gr
from refinenet import RefineNet
# craft, refine ๋ชจ๋ธ ๋ถ๋ฌ์ค๋ ์ฝ๋
def copyStateDict(state_dict):
if list(state_dict.keys())[0].startswith("module"):
start_idx = 1
else:
start_idx = 0
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = ".".join(k.split(".")[start_idx:])
new_state_dict[name] = v
return new_state_dict
def str2bool(v):
return v.lower() in ("yes", "y", "true", "t", "1")
parser = argparse.ArgumentParser(description='CRAFT Text Detection')
parser.add_argument('--trained_model', default='weights/craft_mlt_25k.pth', type=str, help='์ฌ์ ํ์ต craft ๋ชจ๋ธ')
parser.add_argument('--text_threshold', default=0.7, type=float, help='text confidence threshold')
parser.add_argument('--low_text', default=0.4, type=float, help='text low-bound score')
parser.add_argument('--link_threshold', default=0.4, type=float, help='link confidence threshold')
parser.add_argument('--cuda', default=False, type=str2bool, help='Use cuda for inference')
parser.add_argument('--canvas_size', default=1280, type=int, help='image size for inference')
parser.add_argument('--mag_ratio', default=1.5, type=float, help='image magnification ratio')
parser.add_argument('--poly', default=False, action='store_true', help='enable polygon type')
parser.add_argument('--refine', default=True, help='enable link refiner')
parser.add_argument('--image_path', default="input/2.png", help='input image')
parser.add_argument('--refiner_model', default='weights/craft_refiner_CTW1500.pth', type=str, help='pretrained refiner model')
args = parser.parse_args()
# ์๋๋ option
def full_img_masking(full_image,net,refine_net):
reference_image=sr(full_image)
reference_boxes=text_detect(reference_image,net=net,refine_net=refine_net)
boxes=get_box_from_refer(reference_boxes)
for index2,box in enumerate(boxes):
xmin,xmax,ymin,ymax=get_min_max(box)
text_area=full_image[int(ymin):int(ymax),int(xmin):int(xmax),:]
text=recongize(text_area)
label=ner(text)
if label==1:
A=full_image[int(ymin):int(ymax),int(xmin):int(xmax),:]
full_image[int(ymin):int(ymax),int(xmin):int(xmax),:] = gaussian_filter(A, sigma=16)
return full_image
def test_net(net, image, text_threshold, link_threshold, low_text, cuda, poly, refine_net=None):
t0 = time.time()
img_resized, target_ratio, size_heatmap = imgproc.resize_aspect_ratio(image, args.canvas_size, interpolation=cv2.INTER_LINEAR, mag_ratio=args.mag_ratio)
ratio_h = ratio_w = 1 / target_ratio
x = imgproc.normalizeMeanVariance(img_resized)
x = torch.from_numpy(x).permute(2, 0, 1)
x = Variable(x.unsqueeze(0))
if cuda:
x = x.cuda()
with torch.no_grad():
y, feature = net(x)
score_text = y[0,:,:,0].cpu().data.numpy()
score_link = y[0,:,:,1].cpu().data.numpy()
if refine_net is not None:
with torch.no_grad():
y_refiner = refine_net(y, feature)
score_link = y_refiner[0,:,:,0].cpu().data.numpy()
t0 = time.time() - t0
t1 = time.time()
boxes, polys = craft_utils.getDetBoxes(score_text, score_link, text_threshold, link_threshold, low_text, poly)
boxes = craft_utils.adjustResultCoordinates(boxes, ratio_w, ratio_h)
polys = craft_utils.adjustResultCoordinates(polys, ratio_w, ratio_h)
for k in range(len(polys)):
if polys[k] is None: polys[k] = boxes[k]
t1 = time.time() - t1
# render results (optional)
render_img = score_text.copy()
render_img = np.hstack((render_img, score_link))
ret_score_text = imgproc.cvt2HeatmapImg(render_img)
return boxes, polys, ret_score_text
def text_detect(image,net,refine_net):
bboxes, polys, score_text = test_net(net, image, args.text_threshold, args.link_threshold, args.low_text, args.cuda, args.poly, refine_net)
return bboxes
def get_box_from_refer(reference_boxes):
real_boxes=[]
for box in reference_boxes:
real_boxes.append(box//2)
return real_boxes
def get_min_max(box):
xlist=[]
ylist=[]
for coor in box:
xlist.append(coor[0])
ylist.append(coor[1])
return min(xlist),max(xlist),min(ylist),max(ylist)
def main(image_path0):
# 1๋จ๊ณ
# ==> craft ๋ชจ๋ธ๊ณผ refinenet ๋ชจ๋ธ์ ๋ถ๋ฌ์ค๊ณ cuda device ์ ์นํ๋๋ค.
net = CRAFT()
if args.cuda:
net.load_state_dict(copyStateDict(torch.load(args.trained_model)))
if args.cuda:
net = net.cuda()
cudnn.benchmark = False
net.eval()
refine_net = None
if args.refine:
refine_net = RefineNet()
if args.cuda:
refine_net.load_state_dict(copyStateDict(torch.load(args.refiner_model)))
refine_net = refine_net.cuda()
refine_net.eval()
args.poly = True
# 2๋จ๊ณ
# gradio ๋น์นธ์ ์ด๋ฏธ์ง๋ฅผ ๋ฃ๊ณ A ์ ์
๋ ฅ๋ฉ๋๋ค.
A=image_path0
image_list=[]
image_list.append(A)
for k, image_path in enumerate(image_list):
image = imgproc.loadImage(image_path)
if image.shape[2]>3:
image=image[:,:,0:3]
original_image=image
# ์ด๋ฏธ์ง์์ ์ก์ฅ๋ถ๋ถ๋ง dinov2 ๋ชจ๋ธ๋ก segmentation ์ ํฉ๋๋ค.
output=dino_seg(image)
image3=Image.fromarray(output)
image3.save("temporal_mask/mask.png")
# ๋ง์คํฌ์ด๋ฏธ์ง(white pixel, black background)๋ฅผ ๋ง๋ญ๋๋ค.
# ์ ๋ง์คํฌ ์ด๋ฏธ์ง์์ ๊ฐ ๋ฉ์ด๋ฆฌ๋ค(์ก์ฅ์ผ๋ก ์ถ์ )์ด ์ ์ฒด ์ด๋ฏธ์ง๋ด์์ ๋ชํ๋ก์ฐจ์งํ๋์ง ๊ณ์ฐํฉ๋๋ค.
contours_list,percentage_list=mask_percentage("temporal_mask/mask.png")
normal_image_list=[]
small_coordinate_list=[]
original_coordinate_list=[]
#3๋จ๊ณ
sorted_list = sorted(percentage_list, reverse=True)
top_5 = sorted_list[:5]
print("์์ 5๊ฐ ๊ฐ:", top_5)
# percentage list์ ๊ฒฝ์ฐ ์ก์ฅ์ผ๋ก ์ถ์ ๋๋ ๋ญ์น๋ค์ ํผ์ผํธ๋ฅผ ๋ชจ์๋์๊ฒ์ด๊ณ
# contours list๋ ์ด๋ฏธ์ง๋ด์์ ์ก์ฅ์ผ๋ก ์ถ์ ๋๋ ๋ญ์น๋ค์ด ํฌ๋กญ๋์ด์ ์ ๋ ฌ๋ ๋ฆฌ์คํธ์
๋๋ค.
# ์ : percentatge list ์ ์ฒซ๋ฒ์งธ ์์๋ contours list์ ์ฒซ๋ฒ์งธ ์์์ percentage
for index,percentage in enumerate(percentage_list):
if 5<percentage:
# percentage ๊ฐ ์๋ฏธ์ง๋ด์์ 5ํ๋ก ๋๋ ๊ฒ๋ค์ normal list๋ก ํฌํจ๋ฉ๋๋ค.
# normal list์์๋ ์ด๋ฏธ์ง๋ด์์ ์ถฉ๋ถํ ํฐ ๋ญ์น๋ค(์ก์ฅ์ผ๋ก ์ถ์ ) ์ ๋ชจ์๋์์ต๋๋ค.
# 1-5ํ๋ก ์ธ๊ฒ๋ค์ small coordinate list์ ํฌํจ๋๊ณ ๋งค์ฐ ์์ ๋ญ์น๋ก ๊ฐ์ฃผํฉ๋๋ค.
# ๋งค์ฐ์์ ๋ญ์น์ ๊ฒฝ์ฐ zoom in์ ํ์๋ ๋ญ์น(์ก์ฅ์ผ๋ก ์ถ์ )๋ด ๊ธ์๊ฐ ๊ฑฐ์ ๋ณด์ด์ง์์์ ๋ฐ๋ผ์ ๋ญ์น ์ ์ฒด๋ฅผ mosaikํฉ๋๋ค.
# 1ํ๋ก๋ฏธ๋ง ๋ญ์น๋ค์ ์๋ฉธ์ง์ ์ผ์ ๋๋ก ์์ ์๋ตํฉ๋๋ค.
contour=contours_list[index]
x_list=[]
y_list=[]
contour2=list(contour)
for r in contour2:
r2=r[0]
x_list.append(r2[0])
y_list.append(r2[1])
x_min=min(x_list)
y_min=min(y_list)
x_max=max(x_list)
y_max=max(y_list)
original_coordinate_list.append([y_min,y_max,x_min,x_max])
image2=original_image[y_min:y_max,x_min:x_max,:]
normal_image_list.append(image2)
#
elif 1<percentage<5:
contour=contours_list[index]
x_list=[]
y_list=[]
contour2=list(contour)
for r in contour2:
r2=r[0]
x_list.append(r2[0])
y_list.append(r2[1])
x_min=min(x_list)
y_min=min(y_list)
x_max=max(x_list)
y_max=max(y_list)
small_coordinate_list.append([y_min,y_max,x_min,x_max]) #์ก์ฅ 5ํ๋ก๋ฏธ๋ง์ ์ขํ
else:
continue
# 4๋จ๊ณ (๋งค์ฐ์์ ์ก์ฅ)
# small coordinate list์์ ๋งค์ฐ์์ ์ก์ฅ๋ค์ด ๋ชจ์ฌ์ ธ์์ง๋ง list์์ ์์๊ฐ ์์ผ๋ฉด 5๋จ๊ณ๋ก ๋ฐ๋ก๊ฐ๋๋ค.
# ๋ฐ๋ก ๊ฐ์ง์์๊ฒฝ์ฐ(list ์์์ ์ต์ํ๋) mosaik ๋ฅผ ํตํด์ ์ ์ฒด์ด๋ฏธ์ง์์ ์์ ๋ญ์น์ ํด๋นํ๋ ์ขํ๋ค์ ๋ชจ๋ ๋ชจ์์ดํฌํฉ๋๋ค.
if len(small_coordinate_list)>0:
original_image=mosaik(original_image,small_coordinate_list)
else:
pass
# 5๋จ๊ณ (์ด๋์ ๋ ์ฌ์ด์ฆ ์๋ ์ก์ฅ) ==> normal list
# normal image list์์ ์ ์ ํ ํฌ๊ธฐ์ ์ก์ฅ(์ค ํ๋ฉด ๊ธ์ ๋ณด์ด๋) ๋ค์ด ์์ต๋๋ค.
# craft ์
์ฅ์์ text ์์น๋ฅผ return ํ ์ ์๊ฒ๋ ํฌ๋กญ๋ ์ก์ฅ์ esrgan ์ผ๋ก ํ์ง๊ฐ์ ํฉ๋๋ค.
# ํ์ง๊ฐ์ ๋ ์ก์ฅ์ craft์ ๋ฃ์ด์ ์ ํํ๊ฒ text ์ขํ๋ค์ ๋ชจ๋ ๊ตฌํฉ๋๋ค.
# ์ขํ๋ฅผ ๊ตฌํ ๋ ํ์ง ์ข์ ์ก์ฅ์ด๋ฏธ์ง์ ์ขํ๋ฅผ ๊ทธ๋๋ก return ํ์ง ์๊ณ ์๋ณธ ์ก์ฅ์ด๋ฏธ์ง์ ๋ง์ถ์ด์ scale(//2) ํ๊ณ ์ต์ข
์ขํ๋ฅผ ๊ตฌํฉ๋๋ค.
for index,normal_image in enumerate(normal_image_list):
reference_image=sr(normal_image)
reference_boxes=text_detect(reference_image,net=net,refine_net=refine_net)
boxes=get_box_from_refer(reference_boxes)
for index2,box in enumerate(boxes):
xmin,xmax,ymin,ymax=get_min_max(box)
text_area=normal_image[int(ymin):int(ymax),int(xmin):int(xmax),:]
text_area=Image.fromarray(text_area)
os.makedirs("text_area",exist_ok=True)
text_area.save(f"text_area/new_{index2+1}.png")
# 6๋จ๊ณ (text recognize, ner)
# ์ ์ขํ๋ค์ ํตํด์ ์ก์ฅ ๋ด์์ ๋ฐ์ค๋ค์ ํฌ๋กญํฉ๋๋ค.
# ํฌ๋กญ๋ ์ก์ฅ๋ด ๋ถ๋ถ(ํฌ๋กญ๋ ๋ฐ์ค , ์ฆ text ์๋๊ณณ์ผ๋ก ์ถ์ ๋๋๊ณณ) ์ trocr ์๋ฃ์ต๋๋ค.
# trocr์ ์์๋ด์ ์ถ์ ๋๋ text๋ฅผ ๋ณด์ฌ์ค๋๋ค.
# text๋ฅผ ko electra ์๋ฃ์ด์ ํด๋น ์์์์๋ text๊ฐ ๊ฐ์ธ์ ๋ณด์ธ์ง์๋์ง ํ๋ณํฉ๋๋ค.
# ์ก์ฅ๋ด ํด๋น ์์๊ฐ ๊ฐ์ธ์ ๋ณด๋ก(๋ ์ด๋ธ :1) ์ถ์ ๋ ๊ฒฝ์ฐ ๋ชจ์์ดํฌ๋ฅผํฉ๋๋ค.
# ๋ชจ์์ดํฌ๋ผ๊ณ ํ๋ณํ ๊ฒฝ์ฐ ํด๋น์์์ ์ขํ๋ฅผ ์ก์ฅ์ด๋ฏธ์ง์ ๋ง๋ ์ขํ๋ก ๋ณํํ๊ณ ๊ทธ ์ขํ์ ํด๋นํ๋ ๋ถ๋ถ์ ๋ชจ์์ดํฌํฉ๋๋ค.
# ๋ถ๋ถ์ ์ผ๋ก ๋ชจ์์ดํฌ๋ ์ก์ฅ์ด๋ฏธ์ง๋ฅผ ์ ์ฒด์ด๋ฏธ์ง(์ก์ฅ์ ํฌํจํ๋ ์ด๋ฏธ์ง)์ ๋ถ์
๋๋ค.
text=recongize(text_area)
label=ner(text)
with open("output/text_recongnize.txt","a") as recognized:
recognized.writelines(str(index2+1))
recognized.writelines(" ")
recognized.writelines(str(text))
recognized.writelines(" ")
recognized.writelines(str(label))
recognized.writelines("\n")
recognized.close()
print("done")
if label==1:
A=normal_image[int(ymin):int(ymax),int(xmin):int(xmax),:]
normal_image[int(ymin):int(ymax),int(xmin):int(xmax),:] = gaussian_filter(A, sigma=16)
else:
pass
a,b,c,d=original_coordinate_list[index]
original_image[a:b,c:d,:]=normal_image
# ๋ ์ ํ๋ ๋์ด๊ธฐ์ํด์ ์ด๋ฏธ์ง ์ ์ฒด(์ก์ฅ๊ณผ ๋ฐฐ๊ฒฝ ๋๋ค) craft์ ํต์งธ๋ก ๋ฃ๊ธฐ
# ๋จ optional (๋จ์ : infer speed )
#print("full mask start")
#original_image=full_img_masking(original_image,net=net,refine_net=refine_net)
#print("full mask done")
original_image=Image.fromarray(original_image)
original_image.save("output/mosaiked.png")
print("masked complete")
return original_image
# if __name__ == '__main__':
# iface = gr.Interface(
# fn=main,
# inputs=gr.Image(type="filepath", label="Invoice Image"),
# outputs=gr.Image(type="pil", label="Masked Invoice Image"),
# live=True
# )
# iface.launch()
|