def test(testloader, iteration):
net.eval()
with torch.no_grad():
corrects1 = 0
corrects2 = 0
corrects3 = 0
cnt = 0
test_cls_losses = []
test_apn_losses = []
for test_images, test_labels in testloader:
if args.cuda:
test_images = test_images.cuda()
test_labels = test_labels.cuda()
cnt += test_labels.size(0)
# net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(test_images)
preds = []
for i in range(len(test_labels)):
pred = [logit[i][test_labels[i]] for logit in logits]
preds.append(pred)
test_cls_losses = multitask_loss(logits, test_labels)
test_apn_loss = pairwise_ranking_loss(preds)
test_cls_losses.append(sum(test_cls_losses))
test_apn_losses.append(test_apn_loss)
_, predicted1 = torch.max(logits[0], 1)
correct1 = (predicted1 == test_labels).sum()
corrects1 += correct1
_, predicted2 = torch.max(logits[1], 1)
correct2 = (predicted2 == test_labels).sum()
corrects2 += correct2
_, predicted3 = torch.max(logits[2], 1)
correct3 = (predicted3 == test_labels).sum()
corrects3 += correct3
test_cls_losses = torch.stack(test_cls_losses).mean()
test_apn_losses = torch.stack(test_apn_losses).mean()
accuracy1 = corrects1.float() / cnt
accuracy2 = corrects2.float() / cnt
accuracy3 = corrects3.float() / cnt
foo.add_scalar("test_cls_loss", test_cls_losses.item(), iteration + 1)
foo.add_scalar('test_rank_loss', test_apn_losses.item(), iteration + 1)
foo.add_scalar('test_acc1', accuracy1.item(), iteration + 1)
foo.add_scalar('test_acc2', accuracy2.item(), iteration + 1)
foo.add_scalar('test_acc3', accuracy3.item(), iteration + 1)
# logger.scalar_summary('test_cls_loss', test_cls_losses.item(), iteration + 1)
# logger.scalar_summary('test_rank_loss', test_apn_losses.item(), iteration + 1)
# logger.scalar_summary('test_acc1', accuracy1.item(), iteration + 1)
# logger.scalar_summary('test_acc2', accuracy2.item(), iteration + 1)
# logger.scalar_summary('test_acc3', accuracy3.item(), iteration + 1)
print(
" [*] Iter %d || Test accuracy1: %.4f, Test accuracy2: %.4f, Test accuracy3: %.4f"
%
(iteration, accuracy1.item(), accuracy2.item(), accuracy3.item()))
net.train()
def train():
net.train()
conf_loss = 0
loc_loss = 0
print(" [*] Loading dataset...")
batch_iterator = None
trainset = CUB200_loader(os.getcwd() + '/data/CUB_200_2011', split = 'train')
# trainloader = data.DataLoader(trainset, batch_size = 4,
# shuffle = True, collate_fn = trainset.CUB_collate, num_workers = 4)
trainloader = data.DataLoader(trainset, batch_size = args.batch_size,
shuffle = True, collate_fn = trainset.CUB_collate, num_workers = args.num_workers)
testset = CUB200_loader(os.getcwd() + '/data/CUB_200_2011', split = 'test')
# testloader = data.DataLoader(testset, batch_size = 4,
# shuffle = False, collate_fn = testset.CUB_collate, num_workers = 4)
testloader = data.DataLoader(testset, batch_size = args.batch_size,
shuffle = False, collate_fn = testset.CUB_collate, num_workers = args.num_workers)
apn_iter, apn_epoch, apn_steps = pretrainAPN(trainset, trainloader)
cls_iter, cls_epoch, cls_steps = 0, 0, 1
switch_step = 0
old_cls_loss, new_cls_loss = 2, 1
old_apn_loss, new_apn_loss = 2, 1
iteration = 0 # count the both of iteration
epoch_size = len(trainset) // 4
cls_tol = 0
apn_tol = 0
batch_iterator = iter(trainloader)
while ((old_cls_loss - new_cls_loss)**2 > 1e-7) and ((old_apn_loss - new_apn_loss)**2 > 1e-7) and (iteration < 500000):
# until the two type of losses no longer change
print(' [*] Swtich optimize parameters to Class')
while ((cls_tol < 10) and (cls_iter % 5000 != 0)):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if cls_iter % epoch_size == 0:
cls_epoch += 1
if cls_epoch in decay_steps:
cls_steps += 1
adjust_learning_rate(opt1, 0.1, cls_steps, args.lr)
old_cls_loss = new_cls_loss
images, labels = next(batch_iterator)
# images, labels = Variable(images, requires_grad = True), Variable(labels)
images.requires_grad_()
if args.cuda:
images, labels = images.cuda(), labels.cuda()
t0 = time.time()
logits, _, _ = net(images)
opt1.zero_grad()
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
#new_cls_loss = new_cls_losses[0]
new_cls_loss.backward()
opt1.step()
t1 = time.time()
if (old_cls_loss - new_cls_loss)**2 < 1e-6:
cls_tol += 1
else:
cls_tol = 0
logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
logger.scalar_summary('cls_loss1', new_cls_losses[0].item(), iteration + 1)
logger.scalar_summary('cls_loss12', new_cls_losses[1].item(), iteration + 1)
logger.scalar_summary('cls_loss123', new_cls_losses[2].item(), iteration + 1)
iteration += 1
cls_iter += 1
if (cls_iter % 20) == 0:
print(" [*] cls_epoch[%d], Iter %d || cls_iter %d || cls_loss: %.4f || Timer: %.4fsec"%(cls_epoch, iteration, cls_iter, new_cls_loss.item(), (t1 - t0)))
images, labels = next(batch_iterator)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
logits, _, _ = net(images)
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
#cls_iter += 1
test(testloader, iteration)
#continue
print(' [*] Swtich optimize parameters to APN')
switch_step += 1
while ((apn_tol < 10) and apn_iter % 5000 != 0):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if apn_iter % epoch_size == 0:
apn_epoch += 1
if apn_epoch in decay_steps:
apn_steps += 1
adjust_learning_rate(opt2, 0.1, apn_steps, args.lr)
old_apn_loss = new_apn_loss
images, labels = next(batch_iterator)
# images, labels = Variable(images, requires_grad = True), Variable(labels)
images.requires_grad_()
if args.cuda:
images, labels = images.cuda(), labels.cuda()
t0 = time.time()
logits, _, _ = net(images)
opt2.zero_grad()
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
new_apn_loss.backward()
opt2.step()
t1 = time.time()
if (old_apn_loss - new_apn_loss)**2 < 1e-6:
apn_tol += 1
else:
apn_tol = 0
logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
apn_iter += 1
if (apn_iter % 20) == 0:
print(" [*] apn_epoch[%d], Iter %d || apn_iter %d || apn_loss: %.4f || Timer: %.4fsec"%(apn_epoch, iteration, apn_iter, new_apn_loss.item(), (t1 - t0)))
switch_step += 1
images, labels = next(batch_iterator)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
iteration += 1
cls_iter += 1
apn_iter += 1
test(testloader, iteration)
def train():
net.train()
conf_loss = 0
loc_loss = 0
print(" [*] Loading dataset...")
batch_iterator = None
trainset = CUB200_loader.CUB200_loader(os.getcwd() + '/data/CUB_200_2011',
split='train')
#train_sampler = torch.utils.data.distributed.DistributedSampler(trainset)
trainloader = data.DataLoader(trainset,
batch_size=6,
shuffle=True,
collate_fn=trainset.CUB_collate,
num_workers=1)
testset = CUB200_loader.CUB200_loader(os.getcwd() + '/data/CUB_200_2011',
split='test')
#test_sampler = torch.utils.data.distributed.DistributedSampler(testset)
testloader = data.DataLoader(testset,
batch_size=6,
shuffle=False,
collate_fn=testset.CUB_collate,
num_workers=1)
test_sample, _ = next(iter(testloader))
apn_iter, apn_epoch, apn_steps = pretrainAPN(trainset, trainloader)
cls_iter, cls_epoch, cls_steps = 0, 0, 1
switch_step = 0
old_cls_loss, new_cls_loss = 2, 1
old_apn_loss, new_apn_loss = 2, 1
iteration = 0 # count the both of iteration
epoch_size = len(trainset) // 4
cls_tol = 0
apn_tol = 0
batch_iterator = iter(trainloader)
while ((old_cls_loss - new_cls_loss)**2 > 1e-7) and (
(old_apn_loss - new_apn_loss)**2 > 1e-7) and (iteration < 500000):
# until the two type of losses no longer change
print(' [*] Swtich optimize parameters to Class')
while ((cls_tol < 10) and (cls_iter % 5000 != 0)):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if cls_iter % epoch_size == 0:
cls_epoch += 1
if cls_epoch in decay_steps:
cls_steps += 1
adjust_learning_rate(opt1, 0.1, cls_steps, args.lr)
old_cls_loss = new_cls_loss
images, labels = next(batch_iterator)
images, labels = Variable(images,
requires_grad=True), Variable(labels)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
t0 = time.time()
#net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
opt1.zero_grad()
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
#new_cls_loss = new_cls_losses[0]
new_cls_loss.backward()
opt1.step()
t1 = time.time()
if (old_cls_loss - new_cls_loss)**2 < 1e-6:
cls_tol += 1
else:
cls_tol = 0
# logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
# logger.scalar_summary('cls_loss1', new_cls_losses[0].item(), iteration + 1)
# logger.scalar_summary('cls_loss12', new_cls_losses[1].item(), iteration + 1)
# logger.scalar_summary('cls_loss123', new_cls_losses[2].item(), iteration + 1)
iteration += 1
cls_iter += 1
if (cls_iter % 20) == 0:
print(
" [*] cls_epoch[%d], Iter %d || cls_iter %d || cls_loss: %.4f || Timer: %.4fsec"
% (cls_epoch, iteration, cls_iter, new_cls_loss.item(),
(t1 - t0)))
images, labels = next(batch_iterator)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
#net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
# logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
#cls_iter += 1
test(testloader, iteration)
#continue
print(' [*] Swtich optimize parameters to APN')
switch_step += 1
while ((apn_tol < 10) and apn_iter % 5000 != 0):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if apn_iter % epoch_size == 0:
apn_epoch += 1
if apn_epoch in decay_steps:
apn_steps += 1
adjust_learning_rate(opt2, 0.1, apn_steps, args.lr)
old_apn_loss = new_apn_loss
images, labels = next(batch_iterator)
images, labels = Variable(images,
requires_grad=True), Variable(labels)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
t0 = time.time()
#net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images) ##修改
opt2.zero_grad()
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
new_apn_loss.backward()
opt2.step()
t1 = time.time()
if (old_apn_loss - new_apn_loss)**2 < 1e-6:
apn_tol += 1
else:
apn_tol = 0
# logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
apn_iter += 1
if (apn_iter % 20) == 0:
print(
" [*] apn_epoch[%d], Iter %d || apn_iter %d || apn_loss: %.4f || Timer: %.4fsec"
% (apn_epoch, iteration, apn_iter, new_apn_loss.item(),
(t1 - t0)))
switch_step += 1
images, labels = next(batch_iterator)
if args.cuda:
images, labels = images.cuda(), labels.cuda()
#net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
# logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
iteration += 1
cls_iter += 1
apn_iter += 1
test(testloader, iteration)
_, _, _, crops = net(test_sample)
x1, x2 = crops[0].data, crops[1].data
# visualize cropped inputs
# save_img(x1, path=f'samples/iter_{iteration}@2x.jpg', annotation=f'loss = {avg_loss:.7f}, step = {iteration}')
# save_img(x2, path=f'samples/iter_{iteration}@4x.jpg', annotation=f'loss = {avg_loss:.7f}, step = {iteration}')
torch.save(net.state_dict,
'ckpt/RACNN_vgg_CUB200_iter%d.pth' % iteration)
def train():
net.train()
conf_loss = 0
loc_loss = 0
print(" [*] Loading dataset...")
batch_iterator = None
# trainset = CUB200_loader.CUB200_loader(os.getcwd() + '/data/CUB_200_2011', split = 'train')
# #train_sampler = torch.utils.data.distributed.DistributedSampler(trainset)
# trainloader = data.DataLoader(trainset, batch_size = 6,
# shuffle = True, collate_fn = trainset.CUB_collate, num_workers = 1)
# testset = CUB200_loader.CUB200_loader(os.getcwd() + '/data/CUB_200_2011', split = 'test')
# #test_sampler = torch.utils.data.distributed.DistributedSampler(testset)
# testloader = data.DataLoader(testset, batch_size = 6,
# shuffle = False, collate_fn = testset.CUB_collate, num_workers = 1)
# test_sample, _ = next(iter(testloader))
std = 1. / 255.
means = [109.97 / 255., 127.34 / 255., 123.88 / 255.]
transform_train = transforms.Compose([
transforms.Resize(448),
transforms.RandomCrop([448, 448]),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=means, std=[std] * 3)
])
transform_val = transforms.Compose([
transforms.Resize(448),
transforms.CenterCrop(448),
transforms.ToTensor(),
transforms.Normalize(mean=means, std=[std] * 3)
])
trainset = vocLoader.VOCDetection('voc',
year='2007',
image_set='train',
download=False,
transform=transform_train)
# train_sampler = torch.utils.data.distributed.DistributedSampler(trainset)
trainloader = data.DataLoader(trainset,
batch_size=4,
shuffle=True,
num_workers=1,
pin_memory=True)
testset = vocLoader.VOCDetection('voc',
year='2007',
image_set='val',
download=False,
transform=transform_val)
# test_sampler = torch.utils.data.distributed.DistributedSampler(testset)
testloader = data.DataLoader(testset,
batch_size=4,
shuffle=False,
num_workers=1,
pin_memory=True)
test_sample, _ = next(iter(testloader))
apn_iter, apn_epoch, apn_steps = pretrainAPN(trainset, trainloader)
cls_iter, cls_epoch, cls_steps = 0, 0, 1
switch_step = 0
old_cls_loss, new_cls_loss = 2, 1
old_apn_loss, new_apn_loss = 2, 1
iteration = 0 # count the both of iteration
epoch_size = len(trainset) // 4
cls_tol = 0
apn_tol = 0
batch_iterator = iter(trainloader)
doLoad = False
if doLoad:
checkpoint = torch.load('ckpt/RACNN_vgg_voc_iter4982.pth') # 自己指定
net.load_state_dict(checkpoint['state_dict'])
iteration = checkpoint['epoch']
# count = 5201*epoch_init
else:
iteration = 0
while ((old_cls_loss - new_cls_loss)**2 > 1e-7) and (
(old_apn_loss - new_apn_loss)**2 > 1e-7) and (iteration < 500000):
# until the two type of losses no longer change
print(' [*] Swtich optimize parameters to Class')
while ((cls_tol < 10) and (cls_iter % 5000 != 0)):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if cls_iter % epoch_size == 0:
cls_epoch += 1
if cls_epoch in decay_steps:
cls_steps += 1
adjust_learning_rate(opt1, 0.1, cls_steps, args.lr)
old_cls_loss = new_cls_loss
try: # 添加
images, labels = next(batch_iterator)
except StopIteration:
images, labels = Variable(images,
requires_grad=True), Variable(labels)
else:
images, labels = Variable(images,
requires_grad=True), Variable(labels)
if args.cuda and not is_dp:
images = images.cuda()
labels = labels.cuda()
t0 = time.time()
# net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
opt1.zero_grad()
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
# new_cls_loss = new_cls_losses[0]
new_cls_loss.backward()
opt1.step()
t1 = time.time()
if (old_cls_loss - new_cls_loss)**2 < 1e-6:
cls_tol += 1
else:
cls_tol = 0
foo.add_scalar("cls_loss", new_cls_loss.item(), iteration + 1)
foo.add_scalar("cls_loss1", new_cls_losses[0].item(),
iteration + 1)
foo.add_scalar("cls_loss12", new_cls_losses[1].item(),
iteration + 1)
foo.add_scalar("cls_loss123", new_cls_losses[2].item(),
iteration + 1)
# logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
# logger.scalar_summary('cls_loss1', new_cls_losses[0].item(), iteration + 1)
# logger.scalar_summary('cls_loss12', new_cls_losses[1].item(), iteration + 1)
# logger.scalar_summary('cls_loss123', new_cls_losses[2].item(), iteration + 1)
iteration += 1
cls_iter += 1
if (cls_iter % 20) == 0:
print(
" [*] cls_epoch[%d], Iter %d || cls_iter %d || cls_loss: %.4f || Timer: %.4fsec"
% (cls_epoch, iteration, cls_iter, new_cls_loss.item(),
(t1 - t0)))
with torch.no_grad():
try:
images, labels = next(batch_iterator)
except StopIteration:
if args.cuda:
images = images.cuda()
else:
if args.cuda:
images = images.cuda()
labels = labels.cuda()
# net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
foo.add_scalar("rank_loss", new_apn_loss.item(), iteration + 1)
# logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
#cls_iter += 1
test(testloader, iteration)
# continue
print(' [*] Swtich optimize parameters to APN')
switch_step += 1
while ((apn_tol < 10) and apn_iter % 5000 != 0):
if (not batch_iterator) or (iteration % epoch_size == 0):
batch_iterator = iter(trainloader)
if apn_iter % epoch_size == 0:
apn_epoch += 1
if apn_epoch in decay_steps:
apn_steps += 1
adjust_learning_rate(opt2, 0.1, apn_steps, args.lr)
old_apn_loss = new_apn_loss
try:
images, labels = next(batch_iterator)
except StopIteration:
images, labels = Variable(images,
requires_grad=True), Variable(labels)
else:
images, labels = Variable(images,
requires_grad=True), Variable(labels)
if args.cuda and not is_dp:
images = images.cuda()
labels = labels.cuda()
t0 = time.time()
# net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images) # 修改
opt2.zero_grad()
preds = []
for i in range(len(labels)):
pred = [logit[i][labels[i]] for logit in logits]
preds.append(pred)
new_apn_loss = pairwise_ranking_loss(preds)
new_apn_loss.backward()
opt2.step()
t1 = time.time()
if (old_apn_loss - new_apn_loss)**2 < 1e-6:
apn_tol += 1
else:
apn_tol = 0
foo.add_scalar("rank_loss", new_apn_loss.item(), iteration + 1)
# logger.scalar_summary('rank_loss', new_apn_loss.item(), iteration + 1)
iteration += 1
apn_iter += 1
if (apn_iter % 20) == 0:
print(
" [*] apn_epoch[%d], Iter %d || apn_iter %d || apn_loss: %.4f || Timer: %.4fsec"
% (apn_epoch, iteration, apn_iter, new_apn_loss.item(),
(t1 - t0)))
switch_step += 1
with torch.no_grad():
try:
images, labels = next(batch_iterator)
except StopIteration:
if args.cuda and not is_dp:
images = images.cuda()
else:
if args.cuda and not is_dp:
images = images.cuda()
labels = labels.cuda()
# net.to(torch.device("cuda" if torch.cuda.is_available() else "cpu")) ##添加
logits, _, _, _ = net(images)
new_cls_losses = multitask_loss(logits, labels)
new_cls_loss = sum(new_cls_losses)
foo.add_scalar("cls_loss", new_cls_loss.item(), iteration + 1)
# logger.scalar_summary('cls_loss', new_cls_loss.item(), iteration + 1)
iteration += 1
cls_iter += 1
apn_iter += 1
test(testloader, iteration)
if args.cuda and not is_dp:
test_sample = test_sample.cuda()
_, _, _, crops = net(test_sample)
x1, x2 = crops[0].data, crops[1].data
# visualize cropped inputs
# save_img(x1, path=f'samples/iter_{iteration}@2x.jpg', annotation=f'loss = {avg_loss:.7f}, step = {iteration}')
# save_img(x2, path=f'samples/iter_{iteration}@4x.jpg', annotation=f'loss = {avg_loss:.7f}, step = {iteration}')
torch.save(net.state_dict(),
'ckpt/RACNN_vgg_voc_iter%d.pth' % iteration)
