server.py 8.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207
  1. import os
  2. from utils import *
  3. from EmbedNet import *
  4. import torchvision.transforms as transforms
  5. from detectors import S3FD
  6. import argparse
  7. def createParser():
  8. parser = argparse.ArgumentParser(description = "FaceNet");
  9. parser.add_argument('--config', type=str, default=None, help='Config YAML file');
  10. ## Data loader
  11. parser.add_argument('--batch_size', type=int, default=200, help='Batch size, number of classes per batch');
  12. parser.add_argument('--max_img_per_cls', type=int, default=500, help='Maximum number of images per class per epoch');
  13. parser.add_argument('--nDataLoaderThread', type=int, default=5, help='Number of loader threads');
  14. ## Training details
  15. parser.add_argument('--test_interval', type=int, default=5, help='Test and save every [test_interval] epochs');
  16. parser.add_argument('--max_epoch', type=int, default=100, help='Maximum number of epochs');
  17. parser.add_argument('--trainfunc', type=str, default="softmax", help='Loss function');
  18. ## Optimizer
  19. parser.add_argument('--optimizer', type=str, default="adam", help='sgd or adam');
  20. parser.add_argument('--scheduler', type=str, default="steplr", help='Learning rate scheduler');
  21. parser.add_argument('--lr', type=float, default=0.001, help='Learning rate');
  22. parser.add_argument("--lr_decay", type=float, default=0.90, help='Learning rate decay every [test_interval] epochs');
  23. parser.add_argument('--weight_decay', type=float, default=0, help='Weight decay in the optimizer');
  24. ## Loss functions
  25. parser.add_argument("--hard_prob", type=float, default=0.5, help='Hard negative mining probability, otherwise random, only for some loss functions');
  26. parser.add_argument("--hard_rank", type=int, default=10, help='Hard negative mining rank in the batch, only for some loss functions');
  27. parser.add_argument('--margin', type=float, default=0.1, help='Loss margin, only for some loss functions');
  28. parser.add_argument('--scale', type=float, default=30, help='Loss scale, only for some loss functions');
  29. parser.add_argument('--nPerClass', type=int, default=1, help='Number of images per class per batch, only for metric learning based losses');
  30. parser.add_argument('--nClasses', type=int, default=8700, help='Number of classes in the softmax layer, only for softmax-based losses');
  31. ## Load and save
  32. parser.add_argument('--initial_model', type=str, default="./models/amsoft_model.model", help='Initial model weights');
  33. parser.add_argument('--save_path', type=str, default="exps/exp1", help='Path for model and logs');
  34. ## Training and test data
  35. parser.add_argument('--train_path', type=str, default="data/vggface2", help='Absolute path to the train set');
  36. parser.add_argument('--train_ext', type=str, default="jpg", help='Training files extension');
  37. parser.add_argument('--test_path', type=str, default="data/test", help='Absolute path to the test set');
  38. parser.add_argument('--test_list', type=str, default="data/test_list.csv", help='Evaluation list');
  39. ## Model definition
  40. parser.add_argument('--model', type=str, default="ResNet18", help='Name of model definition');
  41. parser.add_argument('--nOut', type=int, default=512, help='Embedding size in the last FC layer');
  42. ## For test only
  43. parser.add_argument('--eval', dest='eval', action='store_true', help='Eval only')
  44. ## For server
  45. parser.add_argument('--server', dest='server', action='store_true', help='Server mode')
  46. parser.add_argument('--feat_save_path', type=str, default='saved_feats', help='Absolute path to the feature')
  47. parser.add_argument('--port', type=int, default=10000, help='Port for the server')
  48. ## Distributed and mixed precision training
  49. parser.add_argument('--mixedprec', dest='mixedprec', action='store_true', help='Enable mixed precision training')
  50. args = parser.parse_args()
  51. return args
  52. def loadParameters(model, path):
  53. state = model.state_dict()
  54. loaded_state = torch.load(path)
  55. for name, param in loaded_state.items():
  56. origname = name;
  57. if name not in state:
  58. if name not in state:
  59. print("%s is not in the model."%origname);
  60. continue;
  61. if state[name].size() != loaded_state[origname].size():
  62. print("Wrong parameter length: %s, model: %s, loaded: %s"%(origname, state[name].size(), loaded_state[origname].size()));
  63. continue;
  64. state[name].copy_(param);
  65. DET = S3FD(device='cuda')
  66. app = Flask(__name__)
  67. args = createParser()
  68. UNKNOWN_THRESHOLD = 0.5
  69. s = EmbedNet(**vars(args)).cuda()
  70. transform = transforms.Compose(
  71. [transforms.ToTensor(),
  72. transforms.Resize(256),
  73. transforms.CenterCrop([224,224]),
  74. transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
  75. # trainer = ModelTrainer(s, **vars(args))
  76. loadParameters(s, args.initial_model)
  77. s.eval()
  78. @app.route('/query', methods=['POST'])
  79. def query():
  80. # unpack the received data
  81. data = pickle.loads(request.get_data())
  82. image = data['img']
  83. image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
  84. bboxes = DET.detect_faces(image_np, conf_th=0.9, scales=[0.5])
  85. if len(bboxes) != 1:
  86. return "fail"
  87. bsi = 100
  88. sx = int((bboxes[0][0]+bboxes[0][2])/2) + bsi
  89. sy = int((bboxes[0][1]+bboxes[0][3])/2) + bsi
  90. ss = int(max((bboxes[0][3]-bboxes[0][1]),(bboxes[0][2]-bboxes[0][0]))/2)
  91. image = numpy.pad(image,((bsi,bsi),(bsi,bsi),(0,0)), 'constant', constant_values=(110,110))
  92. face = image[int(sy-ss):int(sy+ss),int(sx-ss):int(sx+ss)]
  93. face = cv2.resize(face,(240,240))
  94. im1 = Image.fromarray(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
  95. inp1 = transform(im1).cuda()
  96. com_feat = s(inp1).detach().cpu()
  97. files = glob.glob(os.path.join(args.feat_save_path, '*.pt'))
  98. max_score = 0
  99. pname = 'none'
  100. for file in files:
  101. ref_feat = torch.load(file)
  102. score = F.cosine_similarity(ref_feat, com_feat)
  103. if(score>max_score) :
  104. max_score = score.item()
  105. pname = file.split('/')[1].split('.')[0]
  106. print('{} {:.2f}'.format(file,score.item()))
  107. if max_score < UNKNOWN_THRESHOLD:
  108. max_score = 0
  109. pname = "Unknown"
  110. return {
  111. "file":pname,
  112. "score":max_score,
  113. "x1":bboxes[0][0],
  114. "y1":bboxes[0][1],
  115. "x2":bboxes[0][2],
  116. "y2":bboxes[0][3]
  117. }
  118. @app.route('/enroll', methods=['POST'])
  119. def enroll():
  120. # unpack the received data
  121. data = pickle.loads(request.get_data())
  122. iname = data['name']
  123. image = data['img']
  124. image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
  125. bboxes = DET.detect_faces(image_np, conf_th=0.9, scales=[0.5])
  126. bsi = 100
  127. sx = int((bboxes[0][0]+bboxes[0][2])/2) + bsi
  128. sy = int((bboxes[0][1]+bboxes[0][3])/2) + bsi
  129. ss = int(max((bboxes[0][3]-bboxes[0][1]),(bboxes[0][2]-bboxes[0][0]))/2)
  130. image = numpy.pad(image,((bsi,bsi),(bsi,bsi),(0,0)), 'constant', constant_values=(110,110))
  131. face = image[int(sy-ss):int(sy+ss),int(sx-ss):int(sx+ss)]
  132. face = cv2.resize(face,(240,240))
  133. # TO-DO / 2022-08-25
  134. # 0. Client 요구사항 : Enroll 시 종료 시까지 지속해서 사진 전송, 입력값(Name)은 중복없이 고유한 값이라고 가정
  135. # 1. 인물별 폴더에 이미지를 저장
  136. # 2. 이미지 저장 시 중복 방지 처리
  137. # 3. 인물별 폴더의 사진들을 centroid를 통해 feature 추출
  138. if not(os.path.exists(args.feat_save_path)):
  139. os.makedirs(args.feat_save_path)
  140. cv2.imwrite(os.path.join(args.feat_save_path, '{}.jpg'.format(iname)),face)
  141. im1 = Image.fromarray(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
  142. inp1 = transform(im1).cuda()
  143. ref_feat = s(inp1).detach().cpu()
  144. torch.save(ref_feat, os.path.join(args.feat_save_path,'{}.pt'.format(iname)))
  145. return "success"
  146. if __name__ == "__main__":
  147. app.run(host='0.0.0.0', debug=True, port=args.port, threaded=False)