def main():
t = transforms.Compose([transforms.ToPILImage(),
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5])
])
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.eval()
for name, m in model.named_modules():
print(name)
if name == 'model.sa2.sigmoid':
m.register_forward_hook(viz)
img = cv2.imread('1_face_img.jpg')
img = t(img).unsqueeze(0).to(device)
with torch.no_grad():
model(img)
# 计算本个批次内的困难样本数量
num_hard += len(anc_hard_embedding)
# 计算这个epoch内的总三元损失和计算损失所用的困难样本个数
triplet_loss_sum += triplet_loss.item()
attention_loss_sum += attention_loss.item()
# 计算这个epoch里的平均损失
avg_triplet_loss = 0 if (num_hard == 0) else triplet_loss_sum / num_hard
avg_attention_loss = 0 if (num_hard
== 0) else attention_loss_sum / num_hard
avg_loss = avg_triplet_loss + avg_attention_loss
epoch_time_end = time.time()
# 出测试集准确度
print("Validating on TestDataset! ...")
model.eval() # 验证模式
with torch.no_grad(): # 不传梯度了
distances, labels = [], []
progress_bar = enumerate(tqdm(test_dataloader))
for batch_index, (data_a, data_b, label) in progress_bar:
# ‘img1, img2, issame’
# data_a, data_b, label这仨是一批的矩阵
data_a = data_a.cuda()
data_b = data_b.cuda()
label = label.cuda()
output_a, output_b = model(data_a), model(data_b)
distance = l2_distance.forward(output_a, output_b)
# 列表里套矩阵
labels.append(label.cpu().detach().numpy())
from torch.nn.modules.distance import PairwiseDistance
from Losses.Triplet_loss import TripletLoss
from validate_on_LFW import evaluate_lfw
from tqdm import tqdm
from config import config
from Data_loader.Data_loader_facenet import train_dataloader, test_dataloader
from Models.Model_for_facenet import model, optimizer_model, start_epoch, flag_train_multi_gpu
# total_time_start = time.time()
# start_epoch = start_epoch
end_epoch = start_epoch + config['epochs']
l2_distance = PairwiseDistance(2).cuda()
# best_roc_auc = 0
# best_accuracy = 0
model.eval()
with torch.no_grad():
distances, labels = [], []
print("Validating on LFW! ...")
progress_bar = enumerate(tqdm(test_dataloader))
for batch_index, (data_a, data_b, label) in progress_bar:
data_a, data_b, label = data_a.cuda(), data_b.cuda(), label.cuda()
output_a, output_b = model(data_a), model(data_b)
distance = l2_distance.forward(output_a,
output_b) # Euclidean distance
distances.append(distance.cpu().detach().numpy())