start_epoch_=start_epoch,
adam=False,
patience=7)
plt.plot(losses, label='Train losses')
plt.plot(test_losses, label='Test losses')
plt.legend()
plt.savefig(f"losses_{siameseNetwork.name}.png")
plt.clf()
# Fine tune the model
sys.stdout.write(f"Starting fine tune of {base}\n")
ft_lr = lr * 0.01
ft_epochs = -1
ft_start_epoch = 0
siameseNetwork.load(f'./models/triplet/best-{model_name_}.pt')
siameseNetwork.prepare_for_fine_tuning()
model_name_ += '_ft'
losses, test_losses = train(siameseNetwork,
train_loader,
test_loader,
ft_epochs,
ft_lr,
model_name_,
start_epoch_=ft_start_epoch,
adam=False)
plt.plot(losses, label='Train losses')
plt.plot(test_losses, label='Test losses')
plt.legend()
patience=7)
start_epoch = 0
plt.plot(losses, label='Train losses')
plt.plot(test_losses, label='Test losses')
plt.legend()
plt.savefig(f"losses_{siameseNetwork.name}.png")
plt.clf()
# Fine tune the model
print("Starting fine tune of", base)
ft_lr = lr * 0.01
ft_epochs = 30
ft_start_epoch = 0
siameseNetwork.load(f'./models/contrastive/best-{model_name_}.pt')
siameseNetwork.prepare_for_fine_tuning()
model_name_ += '_ft'
losses, test_losses = train(siameseNetwork,
train_loader,
test_loader,
ft_epochs,
ft_lr,
model_name_,
start_epoch_=ft_start_epoch,
adam=False)
plt.plot(losses, label='Train losses')
plt.plot(test_losses, label='Test losses')
plt.legend()
if __name__ == '__main__':
lr = 0.001
batch_size = 32
epochs = 100
start_epoch = 0
train_data = FacesDataset(train=True, validation=False, base='resnet101')
train_loader = DataLoader(train_data, batch_size, False)
test = FacesDataset(train=False, validation=False, base='resnet101')
test_loader = DataLoader(test, batch_size, False)
siamese_network = SiameseNetwork(base='resnet101').cuda()
siamese_network.load('./models/triplet/resnet101.pt')
siamese_network.eval()
predictor_ = LinearPredictor().cuda()
sys.stdout.write('Training Linear predictor:\n')
losses, test_losses, train_accuracies_, test_accuracies_ = train(
siamese_network,
predictor_,
train_loader,
test_loader,
epochs,
lr,
'predictor-linear',
start_epoch_=start_epoch,
adam=False,