server.py 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201
  1. from utils import *
  2. from EmbedNet import *
  3. import torchvision.transforms as transforms
  4. from detectors import S3FD
  5. import argparse
  6. def createParser():
  7. parser = argparse.ArgumentParser(description = "FaceNet");
  8. parser.add_argument('--config', type=str, default=None, help='Config YAML file');
  9. ## Data loader
  10. parser.add_argument('--batch_size', type=int, default=200, help='Batch size, number of classes per batch');
  11. parser.add_argument('--max_img_per_cls', type=int, default=500, help='Maximum number of images per class per epoch');
  12. parser.add_argument('--nDataLoaderThread', type=int, default=5, help='Number of loader threads');
  13. ## Training details
  14. parser.add_argument('--test_interval', type=int, default=5, help='Test and save every [test_interval] epochs');
  15. parser.add_argument('--max_epoch', type=int, default=100, help='Maximum number of epochs');
  16. parser.add_argument('--trainfunc', type=str, default="softmax", help='Loss function');
  17. ## Optimizer
  18. parser.add_argument('--optimizer', type=str, default="adam", help='sgd or adam');
  19. parser.add_argument('--scheduler', type=str, default="steplr", help='Learning rate scheduler');
  20. parser.add_argument('--lr', type=float, default=0.001, help='Learning rate');
  21. parser.add_argument("--lr_decay", type=float, default=0.90, help='Learning rate decay every [test_interval] epochs');
  22. parser.add_argument('--weight_decay', type=float, default=0, help='Weight decay in the optimizer');
  23. ## Loss functions
  24. parser.add_argument("--hard_prob", type=float, default=0.5, help='Hard negative mining probability, otherwise random, only for some loss functions');
  25. parser.add_argument("--hard_rank", type=int, default=10, help='Hard negative mining rank in the batch, only for some loss functions');
  26. parser.add_argument('--margin', type=float, default=0.1, help='Loss margin, only for some loss functions');
  27. parser.add_argument('--scale', type=float, default=30, help='Loss scale, only for some loss functions');
  28. parser.add_argument('--nPerClass', type=int, default=1, help='Number of images per class per batch, only for metric learning based losses');
  29. parser.add_argument('--nClasses', type=int, default=8700, help='Number of classes in the softmax layer, only for softmax-based losses');
  30. ## Load and save
  31. parser.add_argument('--initial_model', type=str, default="./models/amsoft_model.model", help='Initial model weights');
  32. parser.add_argument('--save_path', type=str, default="exps/exp1", help='Path for model and logs');
  33. ## Training and test data
  34. parser.add_argument('--train_path', type=str, default="data/vggface2", help='Absolute path to the train set');
  35. parser.add_argument('--train_ext', type=str, default="jpg", help='Training files extension');
  36. parser.add_argument('--test_path', type=str, default="data/test", help='Absolute path to the test set');
  37. parser.add_argument('--test_list', type=str, default="data/test_list.csv", help='Evaluation list');
  38. ## Model definition
  39. parser.add_argument('--model', type=str, default="ResNet18", help='Name of model definition');
  40. parser.add_argument('--nOut', type=int, default=512, help='Embedding size in the last FC layer');
  41. ## For test only
  42. parser.add_argument('--eval', dest='eval', action='store_true', help='Eval only')
  43. parser.add_argument('--server', dest='server', action='store_true', help='Server mode')
  44. parser.add_argument('--port', type=int, default=10000, help='Port for the server')
  45. ## Distributed and mixed precision training
  46. parser.add_argument('--mixedprec', dest='mixedprec', action='store_true', help='Enable mixed precision training')
  47. args = parser.parse_args()
  48. return args
  49. def loadParameters(model, path):
  50. state = model.state_dict()
  51. loaded_state = torch.load(path)
  52. for name, param in loaded_state.items():
  53. origname = name;
  54. if name not in state:
  55. if name not in state:
  56. print("%s is not in the model."%origname);
  57. continue;
  58. if state[name].size() != loaded_state[origname].size():
  59. print("Wrong parameter length: %s, model: %s, loaded: %s"%(origname, state[name].size(), loaded_state[origname].size()));
  60. continue;
  61. state[name].copy_(param);
  62. DET = S3FD(device='cuda')
  63. app = Flask(__name__)
  64. args = createParser()
  65. UNKNOWN_THRESHOLD = 0.5
  66. s = EmbedNet(**vars(args)).cuda()
  67. transform = transforms.Compose(
  68. [transforms.ToTensor(),
  69. transforms.Resize(256),
  70. transforms.CenterCrop([224,224]),
  71. transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
  72. # trainer = ModelTrainer(s, **vars(args))
  73. loadParameters(s, args.initial_model)
  74. s.eval()
  75. @app.route('/query', methods=['POST'])
  76. def query():
  77. # unpack the received data
  78. data = pickle.loads(request.get_data())
  79. image = data['img']
  80. image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
  81. bboxes = DET.detect_faces(image_np, conf_th=0.9, scales=[0.5])
  82. if len(bboxes) != 1:
  83. return "fail"
  84. bsi = 100
  85. sx = int((bboxes[0][0]+bboxes[0][2])/2) + bsi
  86. sy = int((bboxes[0][1]+bboxes[0][3])/2) + bsi
  87. ss = int(max((bboxes[0][3]-bboxes[0][1]),(bboxes[0][2]-bboxes[0][0]))/2)
  88. image = numpy.pad(image,((bsi,bsi),(bsi,bsi),(0,0)), 'constant', constant_values=(110,110))
  89. face = image[int(sy-ss):int(sy+ss),int(sx-ss):int(sx+ss)]
  90. face = cv2.resize(face,(240,240))
  91. im1 = Image.fromarray(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
  92. inp1 = transform(im1).cuda()
  93. com_feat = s(inp1).detach().cpu()
  94. files = glob.glob('saved_feats/*.pt')
  95. max_score = 0
  96. pname = 'none'
  97. for file in files:
  98. ref_feat = torch.load(file)
  99. score = F.cosine_similarity(ref_feat, com_feat)
  100. if(score>max_score) :
  101. max_score = score.item()
  102. pname = file.split('/')[1].split('.')[0]
  103. print('{} {:.2f}'.format(file,score.item()))
  104. if max_score < UNKNOWN_THRESHOLD:
  105. max_score = 0
  106. pname = "Unknown"
  107. return {
  108. "file":pname,
  109. "score":max_score,
  110. "x1":bboxes[0][0],
  111. "y1":bboxes[0][1],
  112. "x2":bboxes[0][2],
  113. "y2":bboxes[0][3]
  114. }
  115. @app.route('/enroll', methods=['POST'])
  116. def enroll():
  117. # unpack the received data
  118. data = pickle.loads(request.get_data())
  119. iname = data['name']
  120. image = data['img']
  121. image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
  122. bboxes = DET.detect_faces(image_np, conf_th=0.9, scales=[0.5])
  123. bsi = 100
  124. sx = int((bboxes[0][0]+bboxes[0][2])/2) + bsi
  125. sy = int((bboxes[0][1]+bboxes[0][3])/2) + bsi
  126. ss = int(max((bboxes[0][3]-bboxes[0][1]),(bboxes[0][2]-bboxes[0][0]))/2)
  127. image = numpy.pad(image,((bsi,bsi),(bsi,bsi),(0,0)), 'constant', constant_values=(110,110))
  128. face = image[int(sy-ss):int(sy+ss),int(sx-ss):int(sx+ss)]
  129. face = cv2.resize(face,(240,240))
  130. # TO-DO / 2022-08-25
  131. # 0. Client 요구사항 : Enroll 시 종료 시까지 지속해서 사진 전송, 입력값(Name)은 중복없이 고유한 값이라고 가정
  132. # 1. 인물별 폴더에 이미지를 저장
  133. # 2. 이미지 저장 시 중복 방지 처리
  134. # 3. 인물별 폴더의 사진들을 centroid를 통해 feature 추출
  135. cv2.imwrite('saved_feats/{}.jpg'.format(iname),face)
  136. im1 = Image.fromarray(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
  137. inp1 = transform(im1).cuda()
  138. ref_feat = s(inp1).detach().cpu()
  139. torch.save(ref_feat,'saved_feats/{}.pt'.format(iname))
  140. return "success"
  141. if __name__ == "__main__":
  142. app.run(host='0.0.0.0', debug=True, port=args.port, threaded=False)