def main():
print(f"Train numbers:{len(dataset)}")
# first train run this line
model = AlexNet().to(device)
# Load model
# if device == 'cuda':
# model = torch.load(MODEL_PATH + MODEL_NAME).to(device)
# else:
# model = torch.load(MODEL_PATH + MODEL_NAME, map_location='cpu')
# cast
cast = torch.nn.CrossEntropyLoss().to(device)
# Optimization
optimizer = torch.optim.Adam(model.parameters(),
lr=LEARNING_RATE,
weight_decay=1e-8)
step = 1
for epoch in range(1, NUM_EPOCHS + 1):
model.train()
# cal one epoch time
start = time.time()
for images, labels in dataset_loader:
images = images.to(device)
labels = labels.to(device)
# Forward pass
outputs = model(images)
loss = cast(outputs, labels)
# Backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
print(f"Step [{step * BATCH_SIZE}/{NUM_EPOCHS * len(dataset)}], "
f"Loss: {loss.item():.8f}.")
step += 1
# cal train one epoch time
end = time.time()
print(f"Epoch [{epoch}/{NUM_EPOCHS}], " f"time: {end - start} sec!")
# Save the model checkpoint
torch.save(model, MODEL_PATH + '/' + MODEL_NAME)
print(f"Model save to {MODEL_PATH + '/' + MODEL_NAME}.")
if loop == 0:
# randomly select <init_samples> many samples
selected = random.sample(unlabeled, init_samples)
else:
# select based on entropy
selected = select_fn(unlabeled, n_rec)
# removed selected samples from the pool of unlabeled data
unlabeled = list(set(unlabeled) - set(selected))
# add selected samples to the pool of labeled data
labeled.extend(selected)
# train the model using labeled data
model.train()
for epoch in range(num_epochs):
sampler = SubsetRandomSampler(labeled)
dataloader = DataLoader(
cifar_train,
batch_size=batch_size,
sampler=sampler,
pin_memory=True)
step = 0
for imgs, labels in dataloader:
imgs, labels = imgs.to(device), labels.to(device)
outputs = model(imgs)
loss = loss_fn(outputs, labels)
optimizer.zero_grad()
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") #指定设备
print("using {} device.".format(device))
data_transform = { #数据预处理
"train": transforms.Compose([transforms.RandomResizedCrop(224),# key 为trian 返回这些方法 随机裁剪 224*224
transforms.RandomHorizontalFlip(),#随机反转
transforms.ToTensor(),#转成
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),#标准化处理
"val": transforms.Compose([transforms.Resize((224, 224)), # cannot 224, must (224, 224)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])}
data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # get data root path
image_path = os.path.join(data_root, "data_set", "dog_data") # flower data set path
assert os.path.exists(image_path), "{} path does not exist.".format(image_path)
train_dataset = datasets.ImageFolder(root=os.path.join(image_path, "train"),
transform=data_transform["train"])#数据预处理
train_num = len(train_dataset) #个数
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx #获取名称所对应索引
cla_dict = dict((val, key) for key, val in flower_list.items()) #遍历 key value 对调
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:#生成json 便于打开
json_file.write(json_str)
batch_size = 32
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size, shuffle=True,
num_workers=nw) #加载
validate_dataset = datasets.ImageFolder(root=os.path.join(image_path, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=4, shuffle=False,
num_workers=nw)
print("using {} images for training, {} images fot validation.".format(train_num,
val_num))
# test_data_iter = iter(validate_loader)
# test_image, test_label = test_data_iter.next()
#
# def imshow(img):
# img = img / 2 + 0.5 # unnormalize
# npimg = img.numpy()
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.show()
#
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_classes=5, init_weights=True) #类别5
net.to(device) #网络设备
loss_function = nn.CrossEntropyLoss() #损失函数
# pata = list(net.parameters())
optimizer = optim.Adam(net.parameters(), lr=0.0002) #adam优化器 对象是网络中可训练参数 学习率 自己调参
save_path = './AlexNet.pth' #保存模型路径
best_acc = 0.0
for epoch in range(10):#训练
# train
net.train() #管理神经元失活
running_loss = 0.0 #统计平均损失
t1 = time.perf_counter() #训练时间
for step, data in enumerate(train_loader, start=0): #遍历数据集
images, labels = data #分为图像 标签
optimizer.zero_grad() #清空梯度信息
outputs = net(images.to(device)) #正向传播 指定设备
loss = loss_function(outputs, labels.to(device)) #损失
loss.backward() #反向传播
optimizer.step() #更新结点参数
# print statistics
running_loss += loss.item() #损失累加
# print train process
rate = (step + 1) / len(train_loader) #打印训练进度
a = "*" * int(rate * 50)
b = "." * int((1 - rate) * 50)
print("\rtrain loss: {:^3.0f}%[{}->{}]{:.3f}".format(int(rate * 100), a, b, loss), end="")
print()
print(time.perf_counter()-t1)
# validate
net.eval() #关闭失活
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
for val_data in validate_loader:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1] #最大就是类别
acc += (predict_y == val_labels.to(device)).sum().item() #预测与真实对比 累加
val_accurate = acc / val_num #准确率
if val_accurate > best_acc: #如果准确率大于历史最优
best_acc = val_accurate #更新
torch.save(net.state_dict(), save_path) #保存权重
print('[epoch %d] train_loss: %.3f test_accuracy: %.3f' % #打印信息
(epoch + 1, running_loss / step, val_accurate))
print('Finished Training')
def train(args):
device = torch.device(f"cuda:{args.device_id}")
model = AlexNet(n_cls=100, useLRN=args.useLRN, useDropOut=args.useDropOut)
# model = AlexNet(num_classes= 100)
criterion = nn.CrossEntropyLoss()
model.to(device)
optimizer = Adam(model.parameters(), lr=args.lr)
train_loader, valid_loader = getLoaders(split="train",
batch_size=args.batch_size,
num_workers=args.num_workers,
aug=args.useAug)
train_loss_arr = []
valid_loss_arr = []
valid_acc_arr = []
valid_top5_arr = []
n_iter = 0
best_loss = float('inf')
best_top1_acc = 0
best_top5_acc = 0
for ep in range(args.epoch):
model.train()
for _, (img, label) in tqdm(enumerate(train_loader),
total=len(train_loader)):
img, label = img.to(device), label.to(device)
optimizer.zero_grad()
pred = model(img)
loss = criterion(pred, label)
# loss = model.criterion(pred, label)
loss.backward()
optimizer.step()
train_loss_arr.append(loss.item())
n_iter += 1
model.eval()
ep_valid_loss_arr = []
ep_acc_arr = []
ep_top5_arr = []
with torch.no_grad():
for _, (img, label) in tqdm(enumerate(valid_loader),
total=len(valid_loader)):
img, label = img.to(device), label.to(device)
pred = model(img)
loss = criterion(pred, label)
# loss = model.criterion(pred, label)
acc = utils.top_k_acc(k=1,
pred=pred.detach().cpu().numpy(),
label=label.detach().cpu().numpy())
acc5 = utils.top_k_acc(k=5,
pred=pred.detach().cpu().numpy(),
label=label.detach().cpu().numpy())
ep_acc_arr.append(acc)
ep_top5_arr.append(acc5)
ep_valid_loss_arr.append(loss.item())
valid_loss = np.mean(ep_valid_loss_arr)
valid_acc = np.mean(ep_acc_arr)
valid_top5 = np.mean(ep_top5_arr)
train_loss = np.mean(train_loss_arr[-len(train_loader):])
valid_loss_arr.append(valid_loss)
if valid_loss < best_loss:
best_loss = valid_loss
best_top1_acc = valid_acc
best_top5_acc = valid_top5
model.cpu()
torch.save(model.state_dict(), "best_model.pth")
model.to(device)
if (ep + 1) % 10 == 0:
model.cpu()
torch.save(
{
"model": model.state_dict(),
"optimizer": optimizer.state_dict(),
"train_loss": train_loss_arr,
"valid_loss": valid_loss_arr,
"valid_acc": valid_acc_arr,
"valid_top5": valid_top5_arr,
"best_loss": best_loss,
"ep": ep,
"n_iter": n_iter,
}, "model_checkpoint.pth")
model.to(device)
print(
f"[{ep}, {n_iter}] train: {train_loss:.4f}, valid: {valid_loss:.4f}, acc: {valid_acc:.4f}, top5: {valid_top5:.4f}"
)
with open("exp_result.txt", "a+") as f:
f.write(
f"{args}, loss: {best_loss:.4f}, top1: {best_top1_acc*100:.1f}, top5: {best_top5_acc*100:.1f}\n"
)
def train(args):
# the number of N way, K shot images
k = args.nway * args.kshot
""" TODO 1.a """
" Make your own model for Few-shot Classification in 'model.py' file."
# model setting
model = AlexNet()
model.cuda()
""" TODO 1.a END """
# pretrained model load
if args.restore_ckpt is not None:
state_dict = torch.load(args.restore_ckpt)
model.load_state_dict(state_dict)
model = torch.nn.DataParallel(model,
device_ids=range(torch.cuda.device_count()))
if args.test_mode == 1:
Test_phase(model, args, k)
else:
# Train data loading
dataset = Dataset(args.dpath, state='train')
train_sampler = Train_Sampler(dataset._labels,
n_way=args.nway,
k_shot=args.kshot,
query=args.query)
data_loader = DataLoader(dataset=dataset,
batch_sampler=train_sampler,
num_workers=8,
pin_memory=True)
# Validation data loading
val_dataset = Dataset(args.dpath, state='val')
val_sampler = Sampler(val_dataset._labels,
n_way=args.nway,
k_shot=args.kshot,
query=args.query)
val_data_loader = DataLoader(dataset=val_dataset,
batch_sampler=val_sampler,
num_workers=8,
pin_memory=True)
""" TODO 1.b (optional) """
" Set an optimizer or scheduler for Few-shot classification (optional) "
# Default optimizer setting
#optimizer = torch.optim.Adam(model.parameters(), lr=1e-5)
optimizer = torch.optim.SGD(model.parameters(),
lr=0.001,
momentum=0.9,
weight_decay=5e-4)
""" TODO 1.b (optional) END """
tl = Averager() # save average loss
ta = Averager() # save average accuracy
# training start
print('train start')
train_correct = 0
train_total = 0
train_loss = 0
test_correct = 0
test_total = 0
test_loss = 0
model.train()
for i in range(args.se + 1, TOTAL):
for episode in data_loader:
optimizer.zero_grad()
data, label = [_ for _ in episode] # load an episode
# split an episode images and labels into shots and query set
# note! data_shot shape is ( nway * kshot, 3, h, w ) not ( kshot * nway, 3, h, w )
# Take care when reshape the data shot
data_shot, data_query = data[:k], data[k:]
label_shot, label_query = label[:k], label[k:]
label_shot = sorted(list(set(label_shot.tolist())))
# convert labels into 0-4 values
label_query = label_query.tolist()
labels = []
for j in range(len(label_query)):
label = label_shot.index(label_query[j])
labels.append(label)
labels = torch.tensor(labels).cuda()
""" TODO 2 ( Same as above TODO 2 ) """
""" Train the model
Input:
data_shot : torch.tensor, shot images, [args.nway * args.kshot, 3, h, w]
be careful when using torch.reshape or .view functions
data_query : torch.tensor, query images, [args.query, 3, h, w]
labels : torch.tensor, labels of query images, [args.query]
output:
loss : torch scalar tensor which used for updating your model
logits : A value to measure accuracy and loss
"""
features_shot = model(data_shot.cuda())
n_sample = int(args.query / args.nway)
features_shot_mean = torch.zeros(args.nway,
features_shot.size(1)).cuda()
for j in range(int(args.nway)):
start = j * args.kshot
end = (j + 1) * args.kshot
features_shot_mean[j] = features_shot[start:end].mean(dim=0)
features_query = model(data_query.cuda())
logits = square_euclidean_metric(features_query,
features_shot_mean)
labels_expanded = labels.view(args.query, 1, 1)
labels_expanded = labels_expanded.expand(args.query, args.nway, 1)
lsoft = F.log_softmax(-logits, dim=1).view(args.kshot, n_sample,
-1)
labels_expanded = labels_expanded.view(lsoft.size())
loss = -lsoft.gather(2, labels_expanded).squeeze().view(-1).mean()
_, pred = lsoft.max(2)
""" TODO 2 END """
acc = count_acc(logits, labels)
tl.add(loss.item())
ta.add(acc)
loss.backward()
optimizer.step()
proto = None
logits = None
loss = None
if (i + 1) % PRINT_FREQ == 0:
print('train {}, loss={:.4f} acc={:.4f}'.format(
i + 1, tl.item(), ta.item()))
# initialize loss and accuracy mean
tl = None
ta = None
tl = Averager()
ta = Averager()
# validation start
if (i + 1) % VAL_FREQ == 0:
print('validation start')
model.eval()
with torch.no_grad():
vl = Averager() # save average loss
va = Averager() # save average accuracy
for j in range(VAL_TOTAL):
for episode in val_data_loader:
data, label = [_.cuda() for _ in episode]
data_shot, data_query = data[:k], data[
k:] # load an episode
label_shot, label_query = label[:k], label[k:]
label_shot = sorted(list(set(label_shot.tolist())))
label_query = label_query.tolist()
labels = []
for j in range(len(label_query)):
label = label_shot.index(label_query[j])
labels.append(label)
labels = torch.tensor(labels).cuda()
""" TODO 2 ( Same as above TODO 2 ) """
""" Train the model
Input:
data_shot : torch.tensor, shot images, [args.nway * args.kshot, 3, h, w]
be careful when using torch.reshape or .view functions
data_query : torch.tensor, query images, [args.query, 3, h, w]
labels : torch.tensor, labels of query images, [args.query]
output:
loss : torch scalar tensor which used for updating your model
logits : A value to measure accuracy and loss
"""
optimizer.zero_grad()
data, label = [_.cuda()
for _ in episode] # load an episode
# split an episode images and labels into shots and query set
# note! data_shot shape is ( nway * kshot, 3, h, w ) not ( kshot * nway, 3, h, w )
# Take care when reshape the data shot
data_shot, data_query = data[:k], data[k:]
label_shot, label_query = label[:k], label[k:]
label_shot = sorted(list(set(label_shot.tolist())))
# convert labels into 0-4 values
label_query = label_query.tolist()
labels = []
for j in range(len(label_query)):
label = label_shot.index(label_query[j])
labels.append(label)
labels = torch.tensor(labels).cuda()
""" TODO 2 ( Same as above TODO 2 ) """
""" Make a loss function and train your own model
Input:
data_shot : torch.tensor, shot images, [args.nway * args.kshot, 3, h, w]
be careful when using torch.reshape or .view functions
(25, 3, 400, 400)
data_query : torch.tensor, query images, [args.query, 3, h, w]
(20, 3, 400, 400)
labels : torch.tensor, labels of query images, [args.query]
(20)
output:
loss : torch scalar tensor which used for updating your model
logits : A value to measure accuracy and loss
"""
features_shot = model(data_shot.cuda())
n_sample = int(args.query / args.nway)
features_shot_mean = torch.zeros(
args.nway, features_shot.size(1)).cuda()
for j in range(int(args.nway)):
start = j * args.kshot
end = (j + 1) * args.kshot
features_shot_mean[j] = features_shot[
start:end].mean(dim=0)
features_query = model(data_query.cuda())
logits = square_euclidean_metric(
features_query, features_shot_mean)
labels_expanded = labels.view(args.query, 1, 1)
labels_expanded = labels_expanded.expand(
args.query, args.nway, 1)
lsoft = F.log_softmax(-logits, dim=1).view(
args.kshot, n_sample, -1)
labels_expanded = labels_expanded.view(lsoft.size())
loss = -lsoft.gather(
2, labels_expanded).squeeze().view(-1).mean()
_, pred = lsoft.max(2)
""" TODO 2 END """
acc = count_acc(logits, labels)
vl.add(loss.item())
va.add(acc)
proto = None
logits = None
loss = None
print('val accuracy mean : %.4f' % va.item())
print('val loss mean : %.4f' % vl.item())
# initialize loss and accuracy mean
vl = None
va = None
vl = Averager()
va = Averager()
if (i + 1) % SAVE_FREQ == 0:
PATH = 'checkpoints/%d_%s.pth' % (i + 1, args.name)
torch.save(model.module.state_dict(), PATH)
print('model saved, iteration : %d' % i)
def main():
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print("using {} device.".format(device))
tbwriter = SummaryWriter(log_dir="./logs")
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(360),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]),
"val":
transforms.Compose([
transforms.Resize(360, 360), # cannot 360, must (360,360)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
data_root = os.path.abspath(os.path.join(os.getcwd(),
"./DATA")) # get data root path
image_path = os.path.join(data_root, "male") # flower data set path
assert os.path.exists(image_path), "{} path does not exist.".format(
image_path)
train_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "train"),
transform=data_transform["train"])
train_num = len(train_dataset)
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=2)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 8
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0,
8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=8,
shuffle=True,
num_workers=nw)
print("using {} images for training, {} images fot validation.".format(
train_num, val_num))
if os.path.exists("./log360.pth"):
net = AlexNet()
#net.load_state_dict(torch.load("./log360.pth", map_location='cuda:2'))
net = torch.load("./log360.pth", 'cpu')
print("continue training")
else:
net = AlexNet(num_classes=3, init_weights=True)
net.to(device)
print("start training anew")
loss_function = nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.0001)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.98)
epochs = 2000
save_path = './AlexNet.pth'
best_acc = 0.0
train_steps = len(train_loader)
#json_path = './class_indices.json'
#json_file = open(json_path, "r")
#class_indict = json.load(json_file)
#model = AlexNet(num_classed=6).to(device)
trainLOSS = [] #save loss
testLOSS = [] #save loss
valACC = [] #save val acc
for epoch in range(epochs):
scheduler.step()
print('LR:{}'.format(scheduler.get_lr()[0]))
# train
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader)
for step, data in enumerate(train_bar):
images, labels = data
optimizer.zero_grad()
outputs = net(images.to(device))
loss = loss_function(outputs, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epochs, loss)
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader, colour='green')
for val_data in val_bar:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_accurate = acc / val_num
tbwriter.add_scalar('train/loss', running_loss / train_steps, epoch)
tbwriter.add_scalar('val/acc', val_accurate, epoch)
trainLOSS.append(running_loss / train_steps)
valACC.append(val_accurate)
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
print(' ')
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
#predict
#weights_path="./AlexNet.pth"
#model.load_state_dict(torch.load(weights_path))
#model.eval()
#with torch.no_grad():
# putput = torch.squeeze(model(img.to(device))).cpu()
# predict = torch.softmax(output, dim=0)
# predict_cla = torch.argmax(predict.numpy)
npLOSS = np.array(trainLOSS)
npVALACC = np.array(valACC)
np.save('./save/loss_epoch_{}'.format(epoch), npLOSS)
np.save('./save/valacc_epoch_{}'.format(epoch), npVALACC)
print('Finished Training')
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]),
"val":
transforms.Compose([
transforms.Resize((224, 224)), # cannot 224, must (224, 224)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
data_root = os.path.abspath(os.path.join(os.getcwd(), "./"))
image_path = os.path.join(data_root, "flower_data")
train_dataset = datasets.ImageFolder(root=image_path + "/train",
transform=data_transform['train'])
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 8
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=0)
validate_dataset = datasets.ImageFolder(root=image_path + "/val",
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=4,
shuffle=True,
num_workers=0)
# test_data_iter = iter(validate_loader)
# test_image, test_label = test_data_iter.next()
#
# def imshow(img):
# img = img / 2 + 0.5 # unnormalize
# npimg = img.numpy()
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.show()
#
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_class=5)
print(net)
net.to(device)
loss_function = nn.CrossEntropyLoss()
# pata = list(net.parameters())
optimizer = optim.Adam(net.parameters(), lr=0.0002)
save_path = './AlexNet.pth'
best_acc = 0.0
for epoch in range(10):
# train
net.train()
running_loss = 0.0
t1 = time.perf_counter()
for step, data in enumerate(train_loader, start=0):
images, labels = data
optimizer.zero_grad()
outputs = net(images.to(device))
loss = loss_function(outputs, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
# print train process
rate = (step + 1) / len(train_loader)
a = "*" * int(rate * 50)
b = "." * int((1 - rate) * 50)
print("\rtrain loss: {:^3.0f}%[{}->{}]{:.3f}".format(
int(rate * 100), a, b, loss),
end="")
print()
print(time.perf_counter() - t1)
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
# 验证过程中不计算损失梯度
with torch.no_grad():
for val_data in validate_loader:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1]
acc += (predict_y == val_labels.to(device)).sum().item()
val_accurate = acc / val_num
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('[epoch %d] train_loss: %.3f test_accuracy: %.3f' %
(epoch + 1, running_loss / step, val_accurate))
print('Finished Training')
# plt.show()
#
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_classes=5, init_weights=True) # 构造一个新模型
#net.load_state_dict(torch.load("./AlexNet.pth"))
net.to(device) # 将数据迁移到GPU（如果有的话）
loss_function = nn.CrossEntropyLoss() # 损失函数
#pata = list(net.parameters())
optimizer = optim.Adam(net.parameters(), lr=0.0002) # 梯度下降优化器
save_path = './AlexNet.pth'
best_acc = 0.0 # 历史最高准确率（每一个epoch得到的模型都有一个新的准确率）
for epoch in range(10):
net.train() # 训练模式开启
running_loss = 0.0
t1 = time.perf_counter()
for step, data in enumerate(train_loader, start=0):
images, labels = data
optimizer.zero_grad() # 各参数梯度置零，否则梯度会累加
outputs = net(images.to(device)) # 将数据正向过一遍网络，得到一个输出
loss = loss_function(outputs, labels.to(device)) # 计算输出值和期望值的某种误差
loss.backward() # 计算 loss函数 对于各权重参数的偏导数，存入各参数的梯度属性
optimizer.step() # 梯度下降法更新各参数
running_loss += loss.item()
rate = (step + 1) / len(train_loader) # 本epoch的进度
a = "*" * int(rate * 50)
b = "." * int((1 - rate) * 50)
print("\rtrain loss: {:^3.0f}%[{}->{}]{:.3f}".format(
class Solver(object):
def __init__(self, config):
self.model = None
self.lr = config.lr
self.epochs = config.epoch
self.train_batch_size = config.trainBatchSize
self.test_batch_size = config.testBatchSize
self.criterion = None
self.optimizer = None
self.scheduler = None
self.device = None
self.cuda = config.cuda
self.train_loader = None
self.test_loader = None
self.is_board = False
def load_data(self):
train_transform = transforms.Compose(
[transforms.RandomHorizontalFlip(),
transforms.ToTensor()])
test_transform = transforms.Compose([transforms.ToTensor()])
train_set = torchvision.datasets.CIFAR10(
root='/mnt/disk50/datasets/cifar',
train=True,
download=True,
transform=train_transform)
self.train_loader = torch.utils.data.DataLoader(
dataset=train_set, batch_size=self.train_batch_size, shuffle=True)
test_set = torchvision.datasets.CIFAR10(
root='/mnt/disk50/datasets/cifar',
train=False,
download=True,
transform=test_transform)
self.test_loader = torch.utils.data.DataLoader(
dataset=test_set, batch_size=self.test_batch_size, shuffle=False)
def load_model_from_pth(self, model_path):
"""Load the pre-trained model weight
:param model_path:
:return:
"""
checkpoint = torch.load(model_path,
map_location=self.device_name)['model']
# TODO:这里需要具体了解原因在哪里?
checkpoint_parameter_name = list(checkpoint.keys())[0]
model_parameter_name = next(self.model.named_parameters())[0]
is_checkpoint = checkpoint_parameter_name.startswith('module.')
is_model = model_parameter_name.startswith('module.')
if is_checkpoint and not is_model:
# 移除checkpoint模型里面参数
new_parameter_check = OrderedDict()
for key, value in checkpoint.items():
if key.startswith('module.'):
new_parameter_check[key[7:]] = value
self.model.load_state_dict(new_parameter_check)
elif not is_checkpoint and is_model:
# 添加module.参数
new_parameter_dict = OrderedDict()
for key, value in checkpoint.items():
if not key.startswith('module.'):
key = 'module.' + key
new_parameter_dict[key] = value
else:
self.model.load_state_dict(checkpoint)
return self.model
def load_model(self):
if self.cuda:
self.device = torch.device('cuda:0')
cudnn.benchmark = True
else:
self.device = torch.device('cpu')
# self.model = LeNet().to(self.device)
self.model = AlexNet().to(self.device)
self.optimizer = optim.Adam(self.model.parameters(), lr=self.lr)
self.scheduler = optim.lr_scheduler.MultiStepLR(self.optimizer,
milestones=[75, 150],
gamma=0.5)
self.criterion = nn.CrossEntropyLoss().to(self.device)
def train(self, writer=None):
print("train:")
self.model.train()
train_loss = 0
train_correct = 0
total = 0
for batch_num, (data, target) in enumerate(self.train_loader):
data, target = data.to(self.device), target.to(self.device)
self.optimizer.zero_grad()
output = self.model(data)
loss = self.criterion(output, target)
loss.backward()
self.optimizer.step()
train_loss += loss.item()
prediction = torch.max(
output,
1) # second param "1" represents the dimension to be reduced
total += target.size(0)
# train_correct incremented by one if predicted right
train_correct += np.sum(
prediction[1].cpu().numpy() == target.cpu().numpy())
progress_bar(
batch_num, len(self.train_loader),
'Loss: %.4f | Acc: %.3f%% (%d/%d)' %
(train_loss / (batch_num + 1), 100. * train_correct / total,
train_correct, total))
# if not writer:
# writer.add_scalar
return train_loss, train_correct / total
def test(self):
print("test:")
self.model.eval()
test_loss = 0
test_correct = 0
total = 0
start = time.time()
with torch.no_grad():
for batch_num, (data, target) in enumerate(self.test_loader):
data, target = data.to(self.device), target.to(self.device)
output = self.model(data)
loss = self.criterion(output, target)
test_loss += loss.item()
prediction = torch.max(output, 1)
total += target.size(0)
test_correct += np.sum(
prediction[1].cpu().numpy() == target.cpu().numpy())
progress_bar(
batch_num, len(self.test_loader),
'Loss: %.4f | Acc: %.3f%% (%d/%d)' %
(test_loss / (batch_num + 1), 100. * test_correct / total,
test_correct, total))
end = time.time()
time_used = end - start
return test_loss, test_correct / total, time_used
def save(self):
model_out_path = "./best_model_new.pkl"
torch.save(self.model.state_dict(), model_out_path)
print("Checkpoint saved to {}".format(model_out_path))
def run(self):
self.load_data()
self.load_model()
# for k, v in self.model.state_dict():
# print('layer{}'.k)
# print(v)
accuracy = 0
writer = SummaryWriter()
for epoch in range(1, self.epochs + 1):
self.scheduler.step(epoch)
print("\n===> epoch: %d/200" % epoch)
train_loss, train_acc = self.train()
test_loss, test_acc = self.test()
# writer.add_scalar('loss_group',{'train_loss':train_loss.numpy(),
# 'test_loss':test_loss.numpy()},epoch)
# writer.add_scalar('acc_group',{'train_acc':train_acc.numpy(),
# 'test_acc':test_acc.numpy()}, epoch)
if test_acc > accuracy:
accuracy = test_acc
self.save()
elif epoch == self.epochs:
print("===> BEST ACC. PERFORMANCE: %.3f%%" % (accuracy * 100))
self.save()
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
batch_size = 16
epochs = 20
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
"val":
transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
}
data_root = os.path.abspath(os.path.join(os.getcwd(),
".")) # get data root path
image_path = os.path.join(data_root, "data_set",
"flower_data") # flower data set path
assert os.path.exists(image_path), "{} path does not exist.".format(
image_path)
train_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "train"),
transform=data_transform["train"])
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0,
8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=nw)
print("using {} images for training, {} images for validation.".format(
train_num, val_num))
# create model
net = AlexNet(num_classes=5)
net.to(device)
# define loss function
loss_function = nn.CrossEntropyLoss()
# construct an optimizer
params = [p for p in net.parameters() if p.requires_grad]
optimizer = optim.Adam(params, lr=0.0001)
best_acc = 0.0
save_path = 'weights/alexnet.pth'
train_steps = len(train_loader)
for epoch in range(epochs):
# train
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader)
for step, data in enumerate(train_bar):
images, labels = data
optimizer.zero_grad()
logits = net(images.to(device))
loss = loss_function(logits, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epochs, loss)
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader)
for val_data in val_bar:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
# loss = loss_function(outputs, test_labels)
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_bar.desc = "valid epoch[{}/{}]".format(epoch + 1, epochs)
val_accurate = acc / val_num
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('Finished Training')
net = AlexNet(num_classes=5, init_weights=True)
net.to(device)
loss_function = nn.CrossEntropyLoss()
# pata = list(net.parameters())
# lr: learning_rate
optimizer = optim.Adam(net.parameters(), lr=0.0002)
save_path = r'D:\Document\GitHub\deep-learning-for-image-processing\pytorch_classification\Test2_alexnet\AlexNet.pth'
best_acc = 0.0
epo = 20
for epoch in range(epo):
# train
net.train()
running_loss = 0.0
t1 = time.perf_counter()
#从train_loader,加载一个batch
for step, data in enumerate(train_loader, start=0):
images, labels = data
optimizer.zero_grad()
outputs = net(images.to(device))
loss = loss_function(outputs, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
# print train process
rate = (step + 1) / len(train_loader)
14: [['solarize', 0.5, 2], ['invert', 0, 0.3]],
15: [['equalize', 0.2, 0], ['autocontrast', 0.6, 0]],
16: [['equalize', 0.2, 8], ['equalize', 0.6, 4]],
17: [['color', 0.9, 9], ['equalize', 0.6, 6]],
18: [['autocontrast', 0.8, 4], ['solarize', 0.2, 8]],
19: [['brightness', 0.1, 3], ['color', 0.7, 0]],
20: [['solarize', 0.4, 5], ['autocontrast', 0.9, 3]],
21: [['translatey', 0.9, 9], ['translatey', 0.7, 9]],
22: [['autocontrast', 0.9, 2], ['solarize', 0.8, 3]],
23: [['equalize', 0.8, 8], ['invert', 0.1, 3]],
24: [['translatey', 0.7, 9], ['autocontrast', 0.9, 1]]
}
# 开始进行训练和测试，训练一轮，测试一轮
for epoch in range(60):
# train
net.train() # 训练过程中，使用之前定义网络中的dropout
running_loss = 0.0
t1 = time.perf_counter()
# preprocess all pics
if not os.path.exists(image_path_cifar + "/train" + str(epoch)):
os.mkdir(image_path_cifar + "/train" + str(epoch))
for name in os.listdir(image_path_cifar + "/train"):
for file in os.listdir(image_path_cifar + "/train/" + name):
jpg = Image.open(image_path_cifar + "/train/" + name + '/' + file)
operate_id = random.randint(0, 24)
operation = operate_dict[operate_id]
jpg = locals()[operation[0][0]](jpg, operation[0][1],
operation[0][2])
jpg = locals()[operation[1][0]](jpg, operation[1][1],
operation[1][2])
def main():
# viz = Visdom()
# viz.line([0.], [0.], win='train_loss', opts=dict(title='train loss'))
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor()
]),
"val":
transforms.Compose([
transforms.Resize((224, 224)), # cannot 224, must (224, 224)
transforms.ToTensor()
])
}
data_root = "/home/zhongsy/datasets/dataset/" # get data root path
train_dataset = datasets.ImageFolder(root=os.path.join(data_root, "train"),
transform=data_transform["train"])
# print(train_dataset.imgs)
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 1
# number of workers
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8])
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=32,
shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(
data_root, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=4,
shuffle=False,
num_workers=nw)
print("using {} images for training, {} images for validation.".format(
train_num, val_num))
# test_data_iter = iter(validate_loader)
# test_image, test_label = test_data_iter.next()
#
# def imshow(img):
# img = img / 2 + 0.5 # unnormalize
# npimg = img.numpy()
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.show()
#
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_classes=2, init_weights=True)
net.to(device)
loss_function = nn.CrossEntropyLoss()
# pata = list(net.parameters())
optimizer = optim.Adam(net.parameters(), lr=0.0001)
epochs = 30
save_path = './AlexNet.pt'
best_acc = 0.0
train_steps = len(train_loader)
global_step = 0
for epoch in range(epochs):
# train
epochloss = 100000
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader)
for step, data in enumerate(train_bar):
images, labels = data
# print("label: ", labels, labels.dtype)
optimizer.zero_grad()
outputs = net(images.to(device))
# print("imges: ", images, images.dtype)
# outputs_ = outputs.squeeze()
# print("output__ : ", outputs_)
# outputs_ = outputs.to(torch.float)
loss = loss_function(outputs, labels.to(device))
# loss = loss.to(torch.float)
if epochloss > loss:
epochloss = loss
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epochs, loss)
# viz.line([epochloss.cpu().detach().numpy()], [global_step],
# win='train_loss', update='append')
global_step += 1
print("[ start val ]")
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader)
for val_data in val_bar:
val_images, val_labels = val_data
val_labels.unsqueeze(1)
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1]
# print("prect ;", predict_y)
# outputs = outputs.squeeze()
# print("out_puts: ", outputs)
# a = torch.gt(outputs, 0.5)
# print("a ", a)
# for i, (data, label_) in enumerate(zip(outputs, val_labels)):
# if abs(data-label_) <= 0.5:
# acc += 1
# viz.images(val_images.view(-1, 3, 224, 224), win='x')
# viz.text(str(predict_y.detach().cpu().numpy()),
# win='pred', opts=dict(title='pred'))
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_accurate = acc / val_num
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net, save_path)
print('Finished Training')
def main():
# 设置运行设备
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
# 数据处理
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]),
"val":
transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
# 存放train与val的路径
image_path = '/home/xulei/数据集大本营/5_flower_data/flower_data' # flower data root path
# 若该目录不存在，在报错并终止程序
assert os.path.exists(image_path), "{} path does not exist.".format(
image_path)
# 定义训练数据集
train_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "train"),
transform=data_transform["train"])
# 训练数据集的文件数量
train_num = len(train_dataset)
# flower_list: {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
# cla_dict : {0: 'daisy', 1: 'dandelion', 2: 'roses', 3: 'sunflowers', 4: 'tulips'}
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
# 要输出json格式，需要对json数据进行编码，要用到函数：json.dumps
# indent=4, 的作用是让字典的内容逐行显示，每个key占一行
# json_str :
# '{
# "0": "daisy",
# "1": "dandelion",
# "2": "roses",
# "3": "sunflowers",
# "4": "tulips"
# }'
json_str = json.dumps(cla_dict, indent=4)
with open('class_idices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 128
nw = min(os.cpu_count(), batch_size if batch_size > 1 else 0,
8) # number of workers nw: 8 ?????
print("using {} dataloader workers every process".format(nw))
train_loader = datas.DataLoader(train_dataset,
batch_size,
shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "val"),
transform=data_transform["val"])
# val_num: 364
val_num = len(validate_dataset)
validate_loader = datas.DataLoader(validate_dataset,
batch_size,
shuffle=False,
num_workers=nw)
print("using {} images for trainning, {} images for validation.".format(
train_num, val_num))
net = AlexNet(num_classes=5).to(device)
loss_function = nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.00004)
epoches = 20
save_path = './AlexNet.pth'
best_acc = 0.0
# train_steps : 26 len(train_loader)= training_images_num/batch_size
train_steps = len(train_loader)
for epoch in range(epoches):
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader) # 进度条
for step, data in enumerate(train_bar):
images, labels = data
optimizer.zero_grad()
outputs = net(images.to(device))
loss = loss_function(outputs, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epoches, loss)
# validata
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader) # , colour='green'
for val_data in val_bar:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_accurate = acc / val_num
print('\n[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print("Finshed Training")
def ceal_learning_algorithm(du: DataLoader,
dl: DataLoader,
dtest: DataLoader,
k: int = 1000,
delta_0: float = 0.005,
dr: float = 0.00033,
t: int = 1,
epochs: int = 10,
criteria: str = 'cl',
max_iter: int = 45):
"""
Algorithm1 : Learning algorithm of CEAL.
For simplicity, I used the same notation in the paper.
Parameters
----------
du: DataLoader
Unlabeled samples
dl : DataLoader
labeled samples
dtest : DataLoader
test data
k: int, (default = 1000)
uncertain samples selection
delta_0: float
hight confidence samples selection threshold
dr: float
threshold decay
t: int
fine-tuning interval
epochs: int
criteria: str
max_iter: int
maximum iteration number.
Returns
-------
"""
logger.info('Initial configuration: len(du): {}, len(dl): {} '.format(
len(du.sampler.indices), len(dl.sampler.indices)))
# Create the model
model = AlexNet(n_classes=256, device=None)
# Initialize the model
logger.info('Intialize training the model on `dl` and test on `dtest`')
model.train(epochs=epochs, train_loader=dl, valid_loader=None)
# Evaluate model on dtest
acc = model.evaluate(test_loader=dtest)
print('====> Initial accuracy: {} '.format(acc))
for iteration in range(max_iter):
logger.info('Iteration: {}: run prediction on unlabeled data '
'`du` '.format(iteration))
pred_prob = model.predict(test_loader=du)
# get k uncertain samples
uncert_samp_idx, _ = get_uncertain_samples(pred_prob=pred_prob,
k=k,
criteria=criteria)
# get original indices
uncert_samp_idx = [du.sampler.indices[idx] for idx in uncert_samp_idx]
# add the uncertain samples selected from `du` to the labeled samples
# set `dl`
dl.sampler.indices.extend(uncert_samp_idx)
logger.info(
'Update size of `dl` and `du` by adding uncertain {} samples'
' in `dl`'
' len(dl): {}, len(du) {}'.format(len(uncert_samp_idx),
len(dl.sampler.indices),
len(du.sampler.indices)))
# get high confidence samples `dh`
hcs_idx, hcs_labels = get_high_confidence_samples(pred_prob=pred_prob,
delta=delta_0)
# get the original indices
hcs_idx = [du.sampler.indices[idx] for idx in hcs_idx]
# remove the samples that already selected as uncertain samples.
hcs_idx = [
x for x in hcs_idx
if x not in list(set(uncert_samp_idx) & set(hcs_idx))
]
# add high confidence samples to the labeled set 'dl'
# (1) update the indices
dl.sampler.indices.extend(hcs_idx)
# (2) update the original labels with the pseudo labels.
for idx in range(len(hcs_idx)):
dl.dataset.labels[hcs_idx[idx]] = hcs_labels[idx]
logger.info(
'Update size of `dl` and `du` by adding {} hcs samples in `dl`'
' len(dl): {}, len(du) {}'.format(len(hcs_idx),
len(dl.sampler.indices),
len(du.sampler.indices)))
if iteration % t == 0:
logger.info('Iteration: {} fine-tune the model on dh U dl'.format(
iteration))
model.train(epochs=epochs, train_loader=dl)
# update delta_0
delta_0 = update_threshold(delta=delta_0, dr=dr, t=iteration)
# remove the uncertain samples from the original `du`
logger.info('remove {} uncertain samples from du'.format(
len(uncert_samp_idx)))
for val in uncert_samp_idx:
du.sampler.indices.remove(val)
acc = model.evaluate(test_loader=dtest)
print("Iteration: {}, len(dl): {}, len(du): {},"
" len(dh) {}, acc: {} ".format(iteration,
len(dl.sampler.indices),
len(du.sampler.indices),
len(hcs_idx), acc))
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
#训练过程
net = AlexNet(num_classes=5, init_weights=True) # 实例化网络（输出类型为5，初始化权重）
net.to(device) # 分配网络到指定的设备（GPU/CPU）训练
loss_function = nn.CrossEntropyLoss() # 交叉熵损失
optimizer = optim.Adam(net.parameters(), lr=0.0002) # 优化器（训练参数，学习率）
save_path = './AlexNet.pth'
best_acc = 0.0
for epoch in range(150):
########################################## train ###############################################
net.train() # 训练过程中开启 Dropout
running_loss = 0.0 # 每个 epoch 都会对 running_loss 清零
time_start = time.perf_counter() # 对训练一个 epoch 计时
for step, data in enumerate(train_loader, start=0): # 遍历训练集，step从0开始计算
images, labels = data # 获取训练集的图像和标签
optimizer.zero_grad() # 清除历史梯度
outputs = net(images.to(device)) # 正向传播
loss = loss_function(outputs, labels.to(device)) # 计算损失
loss.backward() # 反向传播
optimizer.step() # 优化器更新参数
running_loss += loss.item()
# 打印训练进度（使训练过程可视化）
rate = (step + 1) / len(
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
data_transform = {
"train":
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]),
"val":
transforms.Compose([
transforms.Resize((224, 224)), # cannot 224, must (224, 224)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
data_root = os.path.abspath(os.path.join(os.getcwd(),
"../..")) # get data root path
image_path = os.path.join(data_root, "data_set",
"flower_data") # flower data set path
assert os.path.exists(image_path), "{} path does not exist.".format(
image_path)
train_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "train"),
transform=data_transform["train"])
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 32
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0,
8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(
image_path, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=4,
shuffle=False,
num_workers=nw)
print("using {} images for training, {} images for validation.".format(
train_num, val_num))
# test_data_iter = iter(validate_loader)
# test_image, test_label = test_data_iter.next()
#
# def imshow(img):
# img = img / 2 + 0.5 # unnormalize
# npimg = img.numpy()
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.show()
#
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_classes=5, init_weights=True)
net.to(device)
loss_function = nn.CrossEntropyLoss()
# pata = list(net.parameters())
optimizer = optim.Adam(net.parameters(), lr=0.0002)
epochs = 10
save_path = './AlexNet.pth'
best_acc = 0.0
train_steps = len(train_loader)
for epoch in range(epochs):
# train
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader)
for step, data in enumerate(train_bar):
images, labels = data
optimizer.zero_grad()
outputs = net(images.to(device))
loss = loss_function(outputs, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epochs, loss)
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader)
for val_data in val_bar:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_accurate = acc / val_num
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('Finished Training')
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else
"cpu") # torch.device规定训练中所使用的设备
print("using {} device.".format(device))
data_transform = { # data_transform数据预处理
"train":
transforms.Compose([
transforms.RandomResizedCrop(224), # 随机裁剪为224*224
transforms.RandomHorizontalFlip(), # 水平方向随机翻转
transforms.ToTensor(), # 转化为tensor
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]), # 标准化处理
"val":
transforms.Compose([
transforms.Resize((224, 224)), # * cannot 224, must (224, 224)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
}
print(os.getcwd())
# data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # get data root path
# 先获取数据集所在的根目录os.getcwd()
# ^ os.getcwd() 返回当前进程的工作目录，并非当前文件所在的目录
# "../.."表示的是上两层目录，这个要看具体的情况，这是一个相对路径的写法
# ^ os.path.join 路径拼接，拼接后得到的就是当前目录的上两级目录
# ^ os.path.abspath() 获取指定文件或目录的绝对路径（完整路径）
data_root = os.path.abspath(os.getcwd())
image_path = os.path.join(data_root, "data_set",
"flower_data") # flower data set path
# 等价于 image_path = data_root + "data_set/flower_data"
# assert os.path.exists(image_path), "{} path does not exist.".format(image_path)
train_dataset = datasets.ImageFolder(
root=os.path.join(image_path, "train"), # 下载数据集 ，"train"表示是训练集数据
transform=data_transform["train"]) # 使用"train"的预处理方式
train_num = len(train_dataset) # 查看训练集有多少张图片
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx # * .class_to_idx 得到分类名称对应的索引
cla_dict = dict(
(val, key) for key, val in flower_list.items()) # * 将刚刚字典的键值对 变为 值键对
# write dict into json file
json_str = json.dumps(cla_dict, indent=4) # 将刚刚的字典变为json形式
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 32
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0,
8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(
train_dataset, # 加载数据集
batch_size=batch_size,
shuffle=True, # 通过batchsize和随机参数从样本中获取一批批数据
num_workers=nw) # wins下num_workers一般设置为0，linux下num_workers设置可以分布式计算
validate_dataset = datasets.ImageFolder(
root=os.path.join(
image_path, "val"
), # root=os.path.join(image_path, "val")等价于 root=image_path+"val"
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(
validate_dataset,
batch_size=batch_size,
shuffle=False, # batch_size=4, shuffle=True,
num_workers=nw)
print("using {} images for training, {} images for validation.".format(
train_num, val_num))
# 下面是查看数据集的demo
# 注意，第60行的batch_size=4, shuffle=True再查看：
# test_data_iter = iter(validate_loader)
# test_image, test_label = test_data_iter.next()
# def imshow(img):
# img = img / 2 + 0.5 # unnormalize
# npimg = img.numpy()
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.show()
# print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
# imshow(utils.make_grid(test_image))
net = AlexNet(num_classes=5, init_weights=True) # 5个类别的花数据集，初始化权重为True
# 实例化模型对象 net
net.to(device) # ^ net.to(device)将网络放入刚刚指定的设备中
loss_function = nn.CrossEntropyLoss() # 定义损失函数，多类别的交叉熵函数
# pata = list(net.parameters()) # 调试所用，查看模型的参数
optimizer = optim.Adam(
net.parameters(),
lr=0.0002) # 定义Adam优化器，优化对象是网络中所有的可训练参数net.parameters(),以及学习了lr=0.0002
epochs = 10
save_path = './AlexNet.pth' # 保存权重的路径
best_acc = 0.0 # 最佳准确率 best_acc，首先初始化为0，后面再更新
train_steps = len(train_loader)
for epoch in range(epochs): # 迭代10次
# * 因为使用了dropout，只在训练中使用，预测中不使用
# train # & 训练阶段
net.train() # 调用net.train()进入训练阶段，同时使用 dropout 方法
running_loss = 0.0 # 统计训练中的平均损失
train_bar = tqdm(train_loader) # 为了统计训练一个epoch所需时间
for step, data in enumerate(train_bar): # 遍历数据集；数据集分为图像和标签
images, labels = data
optimizer.zero_grad() # 梯度清0
outputs = net(
images.to(device)) # 正向传播，图像放入设备中，然后实例化AlexNet的网络net中
loss = loss_function(
outputs, labels.to(device)) # 计算损失，计算预测值与真实值的损失，这里label也要放入设备中
loss.backward() # 反向传播到每一个节点
optimizer.step() # 更新每一个节点的参数
# print statistics
running_loss += loss.item() # 累加loss值
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(
epoch + 1, epochs, loss) # 为了或者训练进度
# validate # & 测试阶段
net.eval() # 调用net.eval() 进入测试阶段，同时关闭 dropout 方法
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad(): # * with torch.no_grad() 禁止参数跟踪：验证中不计算损失梯度
val_bar = tqdm(validate_loader)
for val_data in val_bar:
val_images, val_labels = val_data # 数据划分为图片和对应的标签
outputs = net(
val_images.to(device)) # 放入网络net中得到输出，输出的维度是 [batch， 10]
predict_y = torch.max(
outputs, dim=1
)[1] # 求出输出的第1个维度（dim=1类别维度）max（只关注最大值对应的位置[1]，不关心数值 ），得到预测值 predict_y
acc += torch.eq(predict_y, val_labels.to(
device)).sum().item() # 统计预测正确的个数 # ^ 通过.item()得到相应的数值
# acc += (predict_y == val_labels.to(device)).sum().item() # 等价的
val_accurate = acc / val_num # 累加的准确率除以样本个数，得到平均准确率
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc: # 如果当前准确率大于历史最优准确率
best_acc = val_accurate # 更新
torch.save(net.state_dict(), save_path)
print('Finished Training')
