test_wholeimage_swap_multispecific.py 5.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145
  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. import glob
  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. multisepcific_dir = opt.multisepcific_dir
  35. torch.nn.Module.dump_patches = True
  36. logoclass = watermark_image('./simswaplogo/simswaplogo.png')
  37. model = create_model(opt)
  38. model.eval()
  39. mse = torch.nn.MSELoss().cuda()
  40. spNorm =SpecificNorm()
  41. app = Face_detect_crop(name='antelope', root='./insightface_func/models')
  42. app.prepare(ctx_id= 0, det_thresh=0.6, det_size=(640,640))
  43. # The specific person to be swapped(source)
  44. source_specific_id_nonorm_list = []
  45. source_path = os.path.join(multisepcific_dir,'SRC_*')
  46. source_specific_images_path = sorted(glob.glob(source_path))
  47. for source_specific_image_path in source_specific_images_path:
  48. specific_person_whole = cv2.imread(source_specific_image_path)
  49. specific_person_align_crop, _ = app.get(specific_person_whole,crop_size)
  50. specific_person_align_crop_pil = Image.fromarray(cv2.cvtColor(specific_person_align_crop[0],cv2.COLOR_BGR2RGB))
  51. specific_person = transformer_Arcface(specific_person_align_crop_pil)
  52. specific_person = specific_person.view(-1, specific_person.shape[0], specific_person.shape[1], specific_person.shape[2])
  53. # convert numpy to tensor
  54. specific_person = specific_person.cuda()
  55. #create latent id
  56. specific_person_downsample = F.interpolate(specific_person, scale_factor=0.5)
  57. specific_person_id_nonorm = model.netArc(specific_person_downsample)
  58. source_specific_id_nonorm_list.append(specific_person_id_nonorm.clone())
  59. # The person who provides id information (list)
  60. target_id_norm_list = []
  61. target_path = os.path.join(multisepcific_dir,'DST_*')
  62. target_images_path = sorted(glob.glob(target_path))
  63. for target_image_path in target_images_path:
  64. img_a_whole = cv2.imread(target_image_path)
  65. img_a_align_crop, _ = app.get(img_a_whole,crop_size)
  66. img_a_align_crop_pil = Image.fromarray(cv2.cvtColor(img_a_align_crop[0],cv2.COLOR_BGR2RGB))
  67. img_a = transformer_Arcface(img_a_align_crop_pil)
  68. img_id = img_a.view(-1, img_a.shape[0], img_a.shape[1], img_a.shape[2])
  69. # convert numpy to tensor
  70. img_id = img_id.cuda()
  71. #create latent id
  72. img_id_downsample = F.interpolate(img_id, scale_factor=0.5)
  73. latend_id = model.netArc(img_id_downsample)
  74. latend_id = F.normalize(latend_id, p=2, dim=1)
  75. target_id_norm_list.append(latend_id.clone())
  76. assert len(target_id_norm_list) == len(source_specific_id_nonorm_list), "The number of images in source and target directory must be same !!!"
  77. ############## Forward Pass ######################
  78. pic_b = opt.pic_b_path
  79. img_b_whole = cv2.imread(pic_b)
  80. img_b_align_crop_list, b_mat_list = app.get(img_b_whole,crop_size)
  81. # detect_results = None
  82. swap_result_list = []
  83. id_compare_values = []
  84. b_align_crop_tenor_list = []
  85. for b_align_crop in img_b_align_crop_list:
  86. b_align_crop_tenor = _totensor(cv2.cvtColor(b_align_crop,cv2.COLOR_BGR2RGB))[None,...].cuda()
  87. b_align_crop_tenor_arcnorm = spNorm(b_align_crop_tenor)
  88. b_align_crop_tenor_arcnorm_downsample = F.interpolate(b_align_crop_tenor_arcnorm, scale_factor=0.5)
  89. b_align_crop_id_nonorm = model.netArc(b_align_crop_tenor_arcnorm_downsample)
  90. id_compare_values.append([])
  91. for source_specific_id_nonorm_tmp in source_specific_id_nonorm_list:
  92. id_compare_values[-1].append(mse(b_align_crop_id_nonorm,source_specific_id_nonorm_tmp).detach().cpu().numpy())
  93. b_align_crop_tenor_list.append(b_align_crop_tenor)
  94. id_compare_values_array = np.array(id_compare_values).transpose(1,0)
  95. min_indexs = np.argmin(id_compare_values_array,axis=0)
  96. min_value = np.min(id_compare_values_array,axis=0)
  97. swap_result_list = []
  98. swap_result_matrix_list = []
  99. for tmp_index, min_index in enumerate(min_indexs):
  100. if min_value[tmp_index] < opt.id_thres:
  101. swap_result = model(None, b_align_crop_tenor_list[tmp_index], target_id_norm_list[min_index], None, True)[0]
  102. swap_result_list.append(swap_result)
  103. swap_result_matrix_list.append(b_mat_list[tmp_index])
  104. else:
  105. pass
  106. if len(swap_result_list) !=0:
  107. reverse2wholeimage(swap_result_list, swap_result_matrix_list, crop_size, img_b_whole, logoclass, os.path.join(opt.output_path, 'result_whole_swap_multispecific.jpg'), opt.no_simswaplogo)
  108. print(' ')
  109. print('************ Done ! ************')
  110. else:
  111. print('The people you specified are not found on the picture: {}'.format(pic_b))