def main():
# Views the training images and displays the distance on anchor-negative and anchor-positive
test_display_triplet_distance = True
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
print('\nNumber of Classes:\n{}\n'.format(len(train_dir.classes)))
# instantiate model and initialize weights
model = FaceModel(embedding_size=args.embedding_size,
num_classes=len(train_dir.classes),
pretrained=False)
model.to(device)
triplet_loss = TripletMarginLoss(args.margin)
optimizer = create_optimizer(model, args.lr)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print('=> loading checkpoint {}'.format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
else:
print('=> no checkpoint found at {}'.format(args.resume))
start = args.start_epoch
end = start + args.epochs
for epoch in range(start, end):
print(80 * '=')
print('Epoch [{}/{}]'.format(epoch, end - 1))
time0 = time.time()
own_train(train_loader, model, triplet_loss, optimizer, epoch,
data_size)
print(f' Execution time = {time.time() - time0}')
print(80 * '=')
if test_display_triplet_distance:
display_triplet_distance(model, train_loader,
LOG_DIR + "/train_{}".format(epoch))
print(80 * '=')
time0 = time.time()
own_test(test_loader, model, epoch)
print(f' Execution time = {time.time() - time0}')
print(80 * '=')
if test_display_triplet_distance:
display_triplet_distance_test(model, test_loader,
LOG_DIR + "/test_{}".format(epoch))
def main():
# Views the training images and displays the distance on anchor-negative and anchor-positive
test_display_triplet_distance = True
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
print('\nNumber of Classes:\n{}\n'.format(len(train_dir.classes)))
# instantiate model and initialize weights
model = FaceModel(inceptionresnet_v1,
embedding_size=args.embedding_size,
num_classes=len(train_dir.classes),
pretrained=False)
if args.cuda:
model.cuda()
optimizer = create_optimizer(model, args.lr)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print('=> loading checkpoint {}'.format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
else:
print('=> no checkpoint found at {}'.format(args.resume))
start = args.start_epoch
end = start + args.epochs
para_model = torch.nn.parallel.data_parallel(model)
for epoch in range(start, end):
train(train_loader, para_model, optimizer, epoch)
# test(test_loader, model, epoch)
# do checkpointing
torch.save({
'epoch': epoch + 1,
'state_dict': model.state_dict()
}, '{}/checkpoint_{}.pth'.format(LOG_DIR, epoch))
if test_display_triplet_distance:
display_triplet_distance(model, train_loader,
LOG_DIR + "/train_{}".format(epoch))
# pathC = os.path.join("/home/yinzi/data4/new_train/checkpoints_C/396e4702", 'best.pth.tar')
pathC = os.path.join("/home/yinzi/data4/new_train/checkpoints_C/1daed2c2",
'best.pth.tar')
# 396e4702 带数据增广,单独F1得分为0.87，AB结果送入，F1得分为0.834
# 1daed2c2 使用了修复的crop数据集，无数据增广训练。单独F1得分为0.8551966, AB结果送入。F1得分为0.650
# state = torch.load(path, map_location=device)
stateAB = torch.load(pathAB)
stageC = torch.load(pathC)
stateABC = torch.load(pathABC)
# model1.load_state_dict(stateAB['model1'])
# select_model.load_state_dict(stateAB['select_net'])
# model2.load_state_dict(stageC['model2'])
model1.load_state_dict(stateAB['model1'])
select_model.load_state_dict(stateAB['select_net'])
model2.load_state_dict(stageC['model2'])
#stateABC 0.8088
#StageC 0.69
# Dataset and Dataloader
# Dataset Read_in Part
root_dir = "/data1/yinzi/datas"
parts_root_dir = "/home/yinzi/data3/recroped_parts"
txt_file_names = {
'train': "exemplars.txt",
'val': "tuning.txt",
'test': "testing.txt"
}
def main():
args = parser.parse_args()
cuda = torch.cuda.is_available()
torch.manual_seed(args.seed)
if cuda:
torch.cuda.manual_seed(args.seed)
# 1. dataset
root = args.root
kwargs = {'num_workers': 4, 'pin_memory': True} if cuda else {}
test_transforms = transforms.Compose([
transforms.Scale(96),
transforms.ToTensor(),
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])
])
test_dataset = ImageFolder(root, transform=test_transforms)
test_loader = torch.utils.data.DataLoader(test_dataset,
batch_size=args.test_batch_size,
shuffle=False,
**kwargs)
val_iterator = validation_iterator(test_loader)
# 2. model
#train_dir = FaceDataset(dir='/media/lior/LinuxHDD/datasets/MSCeleb-cleaned',n_triplets=10)
print('construct model')
model = FaceModel(embedding_size=128, num_classes=3367, pretrained=False)
model = model.cuda()
if args.resume:
if os.path.isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
print("=> loaded checkpoint '{}'".format(args.resume))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
# extract feature
print('extracting feature')
embeds = []
labels = []
for data, target in val_iterator:
if cuda:
data, target = data.cuda(), target.cuda(async=True)
data_var = Variable(data, volatile=True)
# compute output
output = model(data_var)
embeds.append(output.data.cpu().numpy())
labels.append(target.cpu().numpy())
embeds = np.vstack(embeds)
labels = np.hstack(labels)
print('embeds shape is ', embeds.shape)
print('labels shape is ', labels.shape)
# prepare dict for display
namedict = dict()
for i in range(10):
namedict[i] = str(i)
visual_feature_space(embeds, labels, len(test_dataset.classes), namedict)
