test_wholeimage_swapspecific.py 4.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136
  1. import cv2
  2. import torch
  3. import fractions
  4. import numpy as np
  5. from PIL import Image
  6. import torch.nn.functional as F
  7. from torchvision import transforms
  8. from models.models import create_model
  9. from options.test_options import TestOptions
  10. from insightface_func.face_detect_crop_multi import Face_detect_crop
  11. from util.reverse2original import reverse2wholeimage
  12. import os
  13. from util.add_watermark import watermark_image
  14. import torch.nn as nn
  15. from util.norm import SpecificNorm
  16. from parsing_model.model import BiSeNet
  17. def lcm(a, b): return abs(a * b) / fractions.gcd(a, b) if a and b else 0
  18. transformer_Arcface = transforms.Compose([
  19. transforms.ToTensor(),
  20. transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
  21. ])
  22. def _totensor(array):
  23. tensor = torch.from_numpy(array)
  24. img = tensor.transpose(0, 1).transpose(0, 2).contiguous()
  25. return img.float().div(255)
  26. def _toarctensor(array):
  27. tensor = torch.from_numpy(array)
  28. img = tensor.transpose(0, 1).transpose(0, 2).contiguous()
  29. return img.float().div(255)
  30. if __name__ == '__main__':
  31. opt = TestOptions().parse()
  32. start_epoch, epoch_iter = 1, 0
  33. crop_size = 224
  34. torch.nn.Module.dump_patches = True
  35. logoclass = watermark_image('./simswaplogo/simswaplogo.png')
  36. model = create_model(opt)
  37. model.eval()
  38. mse = torch.nn.MSELoss().cuda()
  39. spNorm =SpecificNorm()
  40. app = Face_detect_crop(name='antelope', root='./insightface_func/models')
  41. app.prepare(ctx_id= 0, det_thresh=0.6, det_size=(640,640))
  42. pic_a = opt.pic_a_path
  43. pic_specific = opt.pic_specific_path
  44. # The person who provides id information
  45. img_a_whole = cv2.imread(pic_a)
  46. img_a_align_crop, _ = app.get(img_a_whole,crop_size)
  47. img_a_align_crop_pil = Image.fromarray(cv2.cvtColor(img_a_align_crop[0],cv2.COLOR_BGR2RGB))
  48. img_a = transformer_Arcface(img_a_align_crop_pil)
  49. img_id = img_a.view(-1, img_a.shape[0], img_a.shape[1], img_a.shape[2])
  50. # convert numpy to tensor
  51. img_id = img_id.cuda()
  52. #create latent id
  53. img_id_downsample = F.interpolate(img_id, scale_factor=0.5)
  54. latend_id = model.netArc(img_id_downsample)
  55. latend_id = F.normalize(latend_id, p=2, dim=1)
  56. # The specific person to be swapped
  57. specific_person_whole = cv2.imread(pic_specific)
  58. specific_person_align_crop, _ = app.get(specific_person_whole,crop_size)
  59. specific_person_align_crop_pil = Image.fromarray(cv2.cvtColor(specific_person_align_crop[0],cv2.COLOR_BGR2RGB))
  60. specific_person = transformer_Arcface(specific_person_align_crop_pil)
  61. specific_person = specific_person.view(-1, specific_person.shape[0], specific_person.shape[1], specific_person.shape[2])
  62. # convert numpy to tensor
  63. specific_person = specific_person.cuda()
  64. #create latent id
  65. specific_person_downsample = F.interpolate(specific_person, scale_factor=0.5)
  66. specific_person_id_nonorm = model.netArc(specific_person_downsample)
  67. # specific_person_id_norm = F.normalize(specific_person_id_nonorm, p=2, dim=1)
  68. ############## Forward Pass ######################
  69. pic_b = opt.pic_b_path
  70. img_b_whole = cv2.imread(pic_b)
  71. img_b_align_crop_list, b_mat_list = app.get(img_b_whole,crop_size)
  72. # detect_results = None
  73. swap_result_list = []
  74. id_compare_values = []
  75. b_align_crop_tenor_list = []
  76. for b_align_crop in img_b_align_crop_list:
  77. b_align_crop_tenor = _totensor(cv2.cvtColor(b_align_crop,cv2.COLOR_BGR2RGB))[None,...].cuda()
  78. b_align_crop_tenor_arcnorm = spNorm(b_align_crop_tenor)
  79. b_align_crop_tenor_arcnorm_downsample = F.interpolate(b_align_crop_tenor_arcnorm, scale_factor=0.5)
  80. b_align_crop_id_nonorm = model.netArc(b_align_crop_tenor_arcnorm_downsample)
  81. id_compare_values.append(mse(b_align_crop_id_nonorm,specific_person_id_nonorm).detach().cpu().numpy())
  82. b_align_crop_tenor_list.append(b_align_crop_tenor)
  83. id_compare_values_array = np.array(id_compare_values)
  84. min_index = np.argmin(id_compare_values_array)
  85. min_value = id_compare_values_array[min_index]
  86. if opt.use_mask:
  87. n_classes = 19
  88. net = BiSeNet(n_classes=n_classes)
  89. net.cuda()
  90. save_pth = os.path.join('./parsing_model/checkpoint', '79999_iter.pth')
  91. net.load_state_dict(torch.load(save_pth))
  92. net.eval()
  93. else:
  94. net =None
  95. if min_value < opt.id_thres:
  96. swap_result = model(None, b_align_crop_tenor_list[min_index], latend_id, None, True)[0]
  97. reverse2wholeimage([b_align_crop_tenor_list[min_index]], [swap_result], [b_mat_list[min_index]], crop_size, img_b_whole, logoclass, \
  98. os.path.join(opt.output_path, 'result_whole_swapspecific.jpg'), opt.no_simswaplogo,pasring_model =net,use_mask=opt.use_mask, norm = spNorm)
  99. print(' ')
  100. print('************ Done ! ************')
  101. else:
  102. print('The person you specified is not found on the picture: {}'.format(pic_b))