def handle_data():
args = common_args.get_args()
device = torch.device(1)
# Receive data from front-end
print("received")
if 'file1' not in request.files:
# flash('No file part')
# print("test point 1")
# return redirect(request.url)
csvFile = open("store_name.csv", "r")
reader = csv.reader(csvFile)
for item in reader:
print(item)
filename1 = item[0]
filename2 = item[1]
csvFile.close()
else:
file1 = request.files['file1']
file2 = request.files['file2']
if file1.filename == '' or file2.filename == '':
flash('No selected file')
print("test point 2")
return redirect(request.url)
# saved_file1 = file1.filename
# saved_file2 = file2.filename
filename1 = secure_filename(file1.filename)
filename2 = secure_filename(file2.filename)
csvFile = open("store_name.csv", "w")
writer = csv.writer(csvFile)
writer.writerow([filename1, filename2])
csvFile.close()
print(filename1)
print(filename2)
# image_text = "Image 1: " + filename1 + "; Image 2: " + filename2
image_text = filename1 + "; " + filename2
method = int(request.values.get("method"))
resolution1 = int(request.values.get("resolution1"))
resolution2 = int(request.values.get("resolution2"))
# Deal with received data, including images, model and resolution
p = filename1.rsplit('_', 1)
file_path1 = p[0] + '/' + p[0] + '_' + p[1]
p = filename2.rsplit('_', 1)
file_path2 = p[0] + '/' + p[0] + '_' + p[1]
file_path1 = os.path.join(app.config['UPLOAD_FOLDER'], file_path1)
file_path2 = os.path.join(app.config['UPLOAD_FOLDER'], file_path2)
print(file_path1)
print(filename1)
resized_face1 = cv2.imread(file_path1)
resized_face2 = cv2.imread(file_path2)
print("check")
width1, height1, channel1 = resized_face1.shape
width2, height2, channel2 = resized_face2.shape
print("check")
# cv2.resize(face, (int(96 / self.N), int(112 / self.N)))
resized_face1 = cv2.resize(
resized_face1, (int(width1 / resolution1), int(height1 / resolution1)))
print("check")
resized_face1 = cv2.resize(resized_face1, (width1, height1))
print("check")
resized_face2 = cv2.resize(
resized_face2, (int(width2 / resolution2), int(height2 / resolution2)))
resized_face2 = cv2.resize(resized_face2, (width2, height2))
resized_face_path1 = os.path.join(app.config['UPLOAD_FOLDER'],
'resized_face1.jpg')
resized_face_path2 = os.path.join(app.config['UPLOAD_FOLDER'],
'resized_face2.jpg')
cv2.imwrite(resized_face_path1, resized_face1)
cv2.imwrite(resized_face_path2, resized_face2)
aligned_face1 = get_alignedface(file_path1, filename1)
aligned_face2 = get_alignedface(file_path2, filename2)
aligned_face1 = cv2.resize(aligned_face1,
(int(96 / resolution1), int(112 / resolution1)))
aligned_face2 = cv2.resize(aligned_face2,
(int(96 / resolution2), int(112 / resolution2)))
aligned_face1 = cv2.resize(aligned_face1, (96, 112))
aligned_face2 = cv2.resize(aligned_face2, (96, 112))
aligned_face_path1 = os.path.join(app.config['UPLOAD_FOLDER'],
'aligned_face1.jpg')
aligned_face_path2 = os.path.join(app.config['UPLOAD_FOLDER'],
'aligned_face2.jpg')
cv2.imwrite(aligned_face_path1, aligned_face1)
cv2.imwrite(aligned_face_path2, aligned_face2)
# setup model
if method == 1:
model_root1 = '/home/tangjiawei/project/fyp/saved/30000_net_backbone.pth'
model_root2 = '/home/tangjiawei/PycharmProjects/FYP_Face_Verification/saved/1000_lr_spherenet_df7.pth'
net1 = nerual_net.spherenet(20, args.feature_dim, args.use_pool,
args.use_dropout)
net1.load_state_dict(torch.load(model_root1))
net1.to(device)
net1.eval()
net2 = nerual_net.spherenet(20, args.feature_dim, args.use_pool,
args.use_dropout)
net2.load_state_dict(torch.load(model_root2))
net2.to(device)
net2.eval()
feature1 = net1(
face_ToTensor(aligned_face1).to(device).view([1, 3, 112, 96]))
feature2 = net1(
face_ToTensor(aligned_face2).to(device).view([1, 3, 112, 96]))
feature3 = net2(
face_ToTensor(aligned_face1).to(device).view([1, 3, 112, 96]))
feature4 = net2(
face_ToTensor(aligned_face2).to(device).view([1, 3, 112, 96]))
model = "Resnet"
thd1 = 0.29
thd2 = 0.29
if method == 2:
model_root1 = '/home/tangjiawei/project/fyp/saved/ALEX_NET_HR_11_April.pkl'
model_root2 = '/home/tangjiawei/PycharmProjects/RadimoicDeepFeatureExtraction/model/1000_lr_alexnet_df9.pth'
net1 = models.alexnet(pretrained=True)
net1.classifier[6] = torch.nn.Linear(4096, 10559)
net1.load_state_dict(torch.load(model_root1))
net1.to(device)
net1.eval()
net2 = models.alexnet(pretrained=True)
net2.classifier[6] = torch.nn.Linear(4096, 10559)
net2.load_state_dict(torch.load(model_root2))
net2.to(device)
aligned_face1 = cv2.resize(aligned_face1, (224, 224))
aligned_face2 = cv2.resize(aligned_face2, (224, 224))
net2.eval()
feature1 = net1(
face_ToTensor(aligned_face1).to(device).view([1, 3, 224, 224]))
feature2 = net1(
face_ToTensor(aligned_face2).to(device).view([1, 3, 224, 224]))
feature3 = net2(
face_ToTensor(aligned_face1).to(device).view([1, 3, 224, 224]))
feature4 = net2(
face_ToTensor(aligned_face2).to(device).view([1, 3, 224, 224]))
model = "Alexnet"
thd1 = 0.68
thd2 = 0.68
# get features from received images, comparing two images --> same or different
# feature1 = net1(face_ToTensor(aligned_face1).to(device).view([1, 3, 112, 96]))
# feature2 = net1(face_ToTensor(aligned_face2).to(device).view([1, 3, 112, 96]))
print(feature1.shape)
print(feature2.shape)
score1 = torch.nn.CosineSimilarity()(feature1, feature2)
score1 = score1.cpu().detach().numpy().reshape(-1, 1)
score1 = score1.item()
score2 = torch.nn.CosineSimilarity()(feature3, feature4)
score2 = score2.cpu().detach().numpy().reshape(-1, 1)
score2 = score2.item()
print(score1)
print(score2)
if score1 >= thd1:
text_to_html1 = model + " verification result: Same Person"
confindence_text1 = "Score is {:.2f}".format(score1 * 100)
else:
text_to_html1 = model + " verification result: Different Persons"
confindence_text1 = "Score is {:.2f}".format(score1 * 100)
if score2 >= thd2:
text_to_html2 = "LR-" + model + " verification result: Same Person"
confindence_text2 = "Score is {:.2f}".format(score2 * 100)
else:
text_to_html2 = "LR-" + model + " verification result: Different Persons"
confindence_text2 = "Score is {:.2f}".format(score2 * 100)
return render_template('demo.html', input_face1=resized_face_path1, input_face2=resized_face_path2, aligned_face1=aligned_face_path1,\
aligned_face2=aligned_face_path2,Result_test1=text_to_html1, Confidence1=confindence_text1,\
Result_test2=text_to_html2, Confidence2=confindence_text2, image_text=image_text)
logging.basicConfig(filename='FYP_Resnet.log', level=logging.INFO)
hr_model_root = '/home/tangjiawei/PycharmProjects/FYP_Face_Verification/30000_net_backbone_asoftmax_64.pth'
lr_model_root = '/home/tangjiawei/project/fyp/saved/30000_net_backbone.pth'
if __name__ == '__main__':
args = common_args.get_args()
# setup device and dataloader
device = torch.device(1)
print(device)
dataloader = get_loader('webface',
batch_size=128,
N=args.downsampling_factor)
# Setup network Spherenet10_HR
hr_spherenet = nerual_net.spherenet(64, args.feature_dim, args.use_pool,
args.use_dropout)
hr_spherenet.load_state_dict(torch.load(hr_model_root))
hr_spherenet.to(device)
# # Setup network Spherenet20_HR
# num_layers = 20
# hr_spherenet = nerual_net.spherenet(num_layers, args.feature_dim, args.use_pool, args.use_dropout)
# hr_spherenet.load_state_dict(torch.load(model_root))
# hr_spherenet.to(device)
# Setup network Spherenet20_lR
lr_spherenet = nerual_net.spherenet(20, args.feature_dim, args.use_pool,
args.use_dropout)
lr_spherenet.load_state_dict(torch.load(lr_model_root))
lr_spherenet.to(device)
downsampling_factor = 7
args = common_args.get_args()
device = torch.device(1)
print(device)
# # Setup network Spherenet20_lR
# num_layers = 20
# lr_spherenet_test = nerual_net.spherenet(num_layers, args.feature_dim, args.use_pool, args.use_dropout)
# lr_spherenet_test.load_state_dict(
# torch.load('/home/tangjiawei/PycharmProjects/FYP_Face_Verification/saved/2000_lr_spherenet_df9.pth'))
# lr_spherenet_test.to(device)
#
# for i in range(16):
# lfw_verification_for_validate.run(lr_spherenet_test, feature_dim=512, device=device, N=(i + 1))
spherenet_64 = nerual_net.spherenet(20, args.feature_dim, args.use_pool,
args.use_dropout)
spherenet_64.load_state_dict(torch.load(model_root))
spherenet_64.to(device)
logging.info('test sphereface backbone img1_lr img2_hr')
for i in range(16):
lfw_verification.run(spherenet_64,
feature_dim=512,
device=device,
N=(i + 1))
# import logging
# import nerual_net
# import pandas as pd
# import cv2
# import numpy as np
# from matlab_cp2tform import get_similarity_transform_for_cv2
import lfw_verification
import common_args
logging.basicConfig(filename='FYP_Resnet.log', level=logging.INFO)
model_root = '/home/tangjiawei/project/fyp/saved/30000_net_backbone.pth'
if __name__ == '__main__':
args = common_args.get_args()
# setup device and dataloader
device = torch.device(1)
print(device)
dataloader = get_loader('webface', batch_size=256, N=args.downsampling_factor)
# Setup network Spherenet20_HR
num_layers = 20
hr_spherenet = nerual_net.spherenet(num_layers, args.feature_dim, args.use_pool, args.use_dropout)
hr_spherenet.load_state_dict(torch.load(model_root))
hr_spherenet.to(device)
# Setup network Spherenet20_lR
lr_spherenet = nerual_net.spherenet(num_layers, args.feature_dim, args.use_pool, args.use_dropout)
lr_spherenet.load_state_dict(torch.load(model_root))
lr_spherenet.to(device)
# setup criterion and optimizer
criterion_lr = torch.nn.MSELoss()
optimizer = optim.Adam(lr_spherenet.parameters(),
lr=0.0003,
weight_decay=0.0005,
betas=(0.9, 0.999),
amsgrad=True)