test_wholeimage_swapsingle.py 3.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596
  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_single import Face_detect_crop
  11. from util.reverse2original import reverse2wholeimage
  12. import os
  13. from util.add_watermark import watermark_image
  14. from util.norm import SpecificNorm
  15. from parsing_model.model import BiSeNet
  16. def lcm(a, b): return abs(a * b) / fractions.gcd(a, b) if a and b else 0
  17. transformer_Arcface = transforms.Compose([
  18. transforms.ToTensor(),
  19. transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
  20. ])
  21. def _totensor(array):
  22. tensor = torch.from_numpy(array)
  23. img = tensor.transpose(0, 1).transpose(0, 2).contiguous()
  24. return img.float().div(255)
  25. if __name__ == '__main__':
  26. opt = TestOptions().parse()
  27. start_epoch, epoch_iter = 1, 0
  28. crop_size = 224
  29. torch.nn.Module.dump_patches = True
  30. logoclass = watermark_image('./simswaplogo/simswaplogo.png')
  31. model = create_model(opt)
  32. model.eval()
  33. spNorm =SpecificNorm()
  34. app = Face_detect_crop(name='antelope', root='./insightface_func/models')
  35. app.prepare(ctx_id= 0, det_thresh=0.6, det_size=(640,640))
  36. with torch.no_grad():
  37. pic_a = opt.pic_a_path
  38. img_a_whole = cv2.imread(pic_a)
  39. img_a_align_crop, _ = app.get(img_a_whole,crop_size)
  40. img_a_align_crop_pil = Image.fromarray(cv2.cvtColor(img_a_align_crop[0],cv2.COLOR_BGR2RGB))
  41. img_a = transformer_Arcface(img_a_align_crop_pil)
  42. img_id = img_a.view(-1, img_a.shape[0], img_a.shape[1], img_a.shape[2])
  43. # convert numpy to tensor
  44. img_id = img_id.cuda()
  45. #create latent id
  46. img_id_downsample = F.interpolate(img_id, scale_factor=0.5)
  47. latend_id = model.netArc(img_id_downsample)
  48. latend_id = F.normalize(latend_id, p=2, dim=1)
  49. ############## Forward Pass ######################
  50. pic_b = opt.pic_b_path
  51. img_b_whole = cv2.imread(pic_b)
  52. img_b_align_crop_list, b_mat_list = app.get(img_b_whole,crop_size)
  53. # detect_results = None
  54. swap_result_list = []
  55. b_align_crop_tenor_list = []
  56. for b_align_crop in img_b_align_crop_list:
  57. b_align_crop_tenor = _totensor(cv2.cvtColor(b_align_crop,cv2.COLOR_BGR2RGB))[None,...].cuda()
  58. swap_result = model(None, b_align_crop_tenor, latend_id, None, True)[0]
  59. swap_result_list.append(swap_result)
  60. b_align_crop_tenor_list.append(b_align_crop_tenor)
  61. if opt.use_mask:
  62. n_classes = 19
  63. net = BiSeNet(n_classes=n_classes)
  64. net.cuda()
  65. save_pth = os.path.join('./parsing_model/checkpoint', '79999_iter.pth')
  66. net.load_state_dict(torch.load(save_pth))
  67. net.eval()
  68. else:
  69. net =None
  70. reverse2wholeimage(b_align_crop_tenor_list, swap_result_list, b_mat_list, crop_size, img_b_whole, logoclass, \
  71. os.path.join(opt.output_path, 'result_whole_swapsingle.jpg'), opt.no_simswaplogo,pasring_model =net,use_mask=opt.use_mask, norm = spNorm)
  72. print(' ')
  73. print('************ Done ! ************')