def get_rnn_model(params):
if params.net_model == Models.AlexNet:
model_rnn = AlexNet(params)
elif params.net_model == Models.VGGNet16:
model_rnn = VGG16Net(params)
elif params.net_model == Models.ResNet50 or params.net_model == Models.ResNet101:
model_rnn = ResNet(params)
else: # params.net_model == Models.DenseNet121:
model_rnn = DenseNet(params)
return model_rnn
def eval_models(proceed_step):
params = get_recursive_params(proceed_step)
params = init_save_dirs(params)
if not is_initial_params_suitable(params):
return
if params.fusion_levels and not is_suitable_level_fusion(params):
return
if params.load_features and not is_cnn_rnn_features_available(params, cnn=0):
return
if params.data_type != DataTypes.RGBD and not is_cnn_rnn_features_available(params, cnn=1):
return
logfile_name = params.log_dir + proceed_step + '/' + get_timestamp() + '_' + str(params.trial) + '-' + \
params.net_model + '_' + params.data_type + '_split_' + str(params.split_no) + '.log'
init_logger(logfile_name, params)
if params.net_model == Models.AlexNet:
model = AlexNet(params)
elif params.net_model == Models.VGGNet16:
model = VGG16Net(params)
elif params.net_model == Models.ResNet50 or params.net_model == Models.ResNet101:
model = ResNet(params)
elif params.net_model == Models.DenseNet121:
model = DenseNet(params)
else:
print('{}{}Unsupported model selection! Please check your model choice in arguments!{}'
.format(PrForm.BOLD, PrForm.RED, PrForm.END_FORMAT))
return
model.eval()
import torch
from alexnet_model import AlexNet
from resnet_model import resnet34
import matplotlib.pyplot as plt
import numpy as np
# create model
model = AlexNet(num_classes=5)
# model = resnet34(num_classes=5)
# load model weights
model_weight_path = "./AlexNet.pth" # "resNet34.pth"
model.load_state_dict(torch.load(model_weight_path))
print(model)
weights_keys = model.state_dict().keys()
for key in weights_keys:
# remove num_batches_tracked para(in bn)
if "num_batches_tracked" in key:
continue
# [kernel_number, kernel_channel, kernel_height, kernel_width]
weight_t = model.state_dict()[key].numpy()
# read a kernel information
# k = weight_t[0, :, :, :]
# calculate mean, std, min, max
weight_mean = weight_t.mean()
weight_std = weight_t.std(ddof=1)
weight_min = weight_t.min()
weight_max = weight_t.max()
from torchvision import transforms
data_transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
# data_transform = transforms.Compose(
# [transforms.Resize(256),
# transforms.CenterCrop(224),
# transforms.ToTensor(),
# transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
# create model
model = AlexNet(num_classes=5)
# model = resnet34(num_classes=5)
# load model weights
model_weight_path = "./AlexNet.pth" # "./resNet34.pth"
model.load_state_dict(torch.load(model_weight_path))
print(model)
# load image
img = Image.open("../tulip.jpg")
# [N, C, H, W]
img = data_transform(img)
# expand batch dimension
img = torch.unsqueeze(img, dim=0)
# forward
out_put = model(img)
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,脚本位置
image_path = os.path.join(data_root, "data_set") # rock 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)
rock_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in rock_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))
net = AlexNet(num_classes=7, 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')
image_folder = 'D:\Data\cifar-100-python\cifar-100-python'
checkpoint_dir = "./checkpoints"
weight_decay = 1e-5
momentum = 0.9
lr = 0.05
epochs = 100
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if __name__ == "__main__":
# 可视化
writer = SummaryWriter(
comment='_alexnet_go_Adam_lr={}_momentum={}_epochs={}'.format(
lr, momentum, epochs))
# 模型
model = AlexNet(num_classes=100)
model.to('cuda:0')
writer.add_graph(model, torch.randn([256, 3, 224, 224]).cuda())
# 损失
loss_func = torch.nn.CrossEntropyLoss()
# 优化器 SGD(loss func, grad func...)
optimizer = torch.optim.SGD(filter(lambda p: p.requires_grad,
model.parameters()),
weight_decay=weight_decay,
momentum=momentum,
lr=lr)
# optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr=lr, betas=(0.9, 0.99))
# 学习率更新
def run_overall_steps(params):
# "cuda" if torch.cuda.is_available() else "cpu" instead of that I force to use cuda here
device = torch.device("cuda")
logging.info('Using device "{}"'.format(device))
if params.net_model == Models.AlexNet:
model_rnn = AlexNet(params)
elif params.net_model == Models.VGGNet16:
model_rnn = VGG16Net(params)
elif params.net_model == Models.ResNet50 or params.net_model == Models.ResNet101:
model_rnn = ResNet(params)
else: # params.net_model == Models.DenseNet121:
model_rnn = DenseNet(params)
if params.run_mode == OverallModes.FUSION:
process_fusion(model_rnn, params)
else:
if params.run_mode == OverallModes.FINETUNE_MODEL:
save_dir = params.dataset_path + params.features_root + RunSteps.FINE_TUNING + '/'
best_model_file = save_dir + params.net_model + '_' + params.data_type + '_split_' + \
str(params.split_no) + '_best_checkpoint.pth'
num_classes = len(wrgbd51.class_names)
if params.net_model == Models.DenseNet121:
model_ft = models.densenet121()
num_ftrs = model_ft.classifier.in_features
model_ft.classifier = nn.Linear(num_ftrs, num_classes)
elif params.net_model in (Models.ResNet50, Models.ResNet101):
if params.net_model == Models.ResNet50:
model_ft = models.resnet50()
else:
model_ft = models.resnet101()
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, num_classes)
else: # params.net_model == Models.AlexNet or Models.VGGNet16
if params.net_model == Models.AlexNet:
model_ft = models.alexnet()
else:
model_ft = models.vgg16_bn()
num_ftrs = model_ft.classifier[6].in_features
model_ft.classifier[6] = nn.Linear(num_ftrs, num_classes)
try:
checkpoint = torch.load(best_model_file, map_location=device)
model_ft.load_state_dict(checkpoint)
except Exception as e:
print('{}{}Failed to load the finetuned model: {}{}'.format(PrForm.BOLD, PrForm.RED, e,
PrForm.END_FORMAT))
return
elif params.run_mode == OverallModes.FIX_PRETRAIN_MODEL:
if params.net_model == Models.AlexNet:
model_ft = models.alexnet(pretrained=True)
elif params.net_model == Models.VGGNet16:
model_ft = models.vgg16_bn(pretrained=True)
elif params.net_model == Models.ResNet50:
model_ft = models.resnet50(pretrained=True)
elif params.net_model == Models.ResNet101:
model_ft = models.resnet101(pretrained=True)
else: # params.net_model is Models.DenseNet121
model_ft = models.densenet121(pretrained=True)
# Set model to evaluation mode (without this, results will be completely different)
# Remember that you must call model.eval() to set dropout and batch normalization layers
# to evaluation mode before running inference.
model_ft = model_ft.eval()
model_ft = model_ft.to(device)
data_form = get_data_transform(params.data_type)
training_set = WashingtonDataset(params, phase='train', loader=custom_loader, transform=data_form)
train_loader = torch.utils.data.DataLoader(training_set, params.batch_size, shuffle=False)
test_set = WashingtonDataset(params, phase='test', loader=custom_loader, transform=data_form)
test_loader = torch.utils.data.DataLoader(test_set, params.batch_size, shuffle=False)
data_loaders = {'train': train_loader, 'test': test_loader}
for phase in ['train', 'test']:
batch_ind = 0
for inputs, labels, filenames in data_loaders[phase]:
inputs = inputs.to(device)
features = []
for extracted_layer in range(1, 8):
if params.net_model == Models.AlexNet or params.net_model == Models.VGGNet16:
extractor = AlexNetVGG16Extractor(model_ft, extracted_layer, params.net_model)
elif params.net_model == Models.ResNet50 or params.net_model == Models.ResNet101:
extractor = ResNetExtractor(model_ft, extracted_layer, params.net_model)
else: # params.net_model == Models.DenseNet121:
extractor = DenseNet121Extractor(model_ft, extracted_layer)
features.append(extractor(inputs).detach().cpu().clone().numpy())
process_rnn_stage(params, model_rnn, features, labels, filenames, phase, batch_ind)
batch_ind += 1
process_classification_stage(model_rnn, params.run_mode)
# Network params
num_classes = 5
# How often we want to write the tf.summary data to disk
display_step = 100
# Path for tf.summary.FileWriter and to store model checkpoints
filewriter_path = "E:\\shuqian\\luca\\code\\graph"
checkpoint_path = "E:\\shuqian\\luca\\code\\checkpoints"
# TF placeholder for graph input and output
x = tf.placeholder(tf.float32, [batch_size, 64, 64, 3], name='x_images')
y = tf.placeholder(tf.float32, [batch_size, num_classes], name='y_labels')
# Initialize model
model = AlexNet(x, num_classes)
# Link variable to model output
score = model.fc3
pre = model.pre
# List of trainable variables of the layers we want to train
var_list = tf.trainable_variables()
# Op for calculating the loss
with tf.name_scope("cross_ent"):
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=score, labels=y))
costs = []
for var in var_list:
if var.op.name.find(r'weights') > 0:
num_classes = 5
# How often we want to write the tf.summary data to disk
display_step = 300
# Path for tf.summary.FileWriter and to store model checkpoints
filewriter_path = "H:\\shuqian\\amel\\first\\graph"
checkpoint_path = "H:\\shuqian\\amel\\first\\checkpoints"
# TF placeholder for graph input and output
x = tf.placeholder(tf.float32, [batch_size, 252, 252, 3], name='x_images')
y = tf.placeholder(tf.float32, [batch_size, num_classes], name='y_labels')
keep_prob = tf.placeholder(tf.float32, name='prob_rate')
# Initialize model
model = AlexNet(x, keep_prob, num_classes)
# Link variable to model output
score = model.fc3
pre = model.pre
# List of trainable variables of the layers we want to train
var_list = tf.trainable_variables()
# Op for calculating the loss
with tf.name_scope("cross_ent"):
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=score, labels=y))
# Train op
with tf.name_scope("train"):
def main(_):
if FLAGS.training_epoch <= 0:
print('Please specify a positive value for training iteration.')
sys.exit(-1)
if FLAGS.batch_size <= 0:
print('Please specify a positive value for batch size.')
sys.exit(-1)
batch_size = FLAGS.batch_size
print('Loading data...')
training, testing = create_generator(os.path.join(FLAGS.data_dir,
'i1k-extracted/train'),
os.path.join(FLAGS.data_dir,
'i1k-extracted/val'),
batch_size=batch_size)
print('Data loaded!')
display_step = 20
train_size = training.samples
n_classes = training.num_classes
image_size = 227
img_channel = 3
num_epochs = FLAGS.training_epoch
x_flat = tf.placeholder(tf.float32,
(None, image_size * image_size * img_channel))
x_3d = tf.reshape(x_flat,
shape=(tf.shape(x_flat)[0], image_size, image_size,
img_channel))
y = tf.placeholder(tf.float32, [None, n_classes])
keep_prob = tf.placeholder(tf.float32)
model = AlexNet(x_3d, keep_prob=keep_prob, num_classes=n_classes)
model_train = model.fc8
model_prediction = tf.nn.softmax(model_train)
cost = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=model_train, labels=y))
global_step = tf.Variable(0, trainable=False, name='global_step')
lr = tf.train.exponential_decay(0.01,
global_step,
100000,
0.1,
staircase=True)
optimizer = tf.train.MomentumOptimizer(
learning_rate=lr, momentum=0.9).minimize(cost, global_step=global_step)
accuracy, update_op = tf.metrics.accuracy(labels=tf.argmax(y, 1),
predictions=tf.argmax(
model_prediction, 1))
test_accuracy, test_update_op = tf.metrics.accuracy(labels=tf.argmax(y, 1),
predictions=tf.argmax(
model_prediction,
1))
start_time = time.time()
print("Start time is: {}".format(str(start_time)))
with tf.Session() as sess:
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
for step in range(int(num_epochs * train_size) // batch_size):
batch_xs, batch_ys = training.next()
sess.run(optimizer,
feed_dict={
x_3d: batch_xs,
y: batch_ys,
keep_prob: 0.5
})
sess.run(lr)
if step % display_step == 0:
acc_up = sess.run([accuracy, update_op],
feed_dict={
x_3d: batch_xs,
y: batch_ys,
keep_prob: 1.
})
acc = sess.run(accuracy,
feed_dict={
x_3d: batch_xs,
y: batch_ys,
keep_prob: 1.
})
loss = sess.run(cost,
feed_dict={
x_3d: batch_xs,
y: batch_ys,
keep_prob: 1.
})
elapsed_time = time.time() - start_time
print(" Iter " + str(step) + ", Minibatch Loss= " +
"{:.6f}".format(loss) + ", Training Accuracy= " +
"{}".format(acc) + " Elapsed time:" + str(elapsed_time))
stop_time = time.time()
print("Optimization Finished!")
print("Training took: {}".format(stop_time - start_time))
step_test = 1
acc_list = []
while step_test * batch_size < testing.samples:
testing_xs, testing_ys = testing.next()
acc_up = sess.run([test_accuracy, test_update_op],
feed_dict={
x_3d: testing_xs,
y: testing_ys,
keep_prob: 1.
})
acc = sess.run([test_accuracy],
feed_dict={
x_3d: testing_xs,
y: testing_ys,
keep_prob: 1.
})
acc_list.extend(acc)
step_test += 1
# save model using SavedModelBuilder from TF
export_path_base = FLAGS.export_dir
export_path = os.path.join(tf.compat.as_bytes(export_path_base),
tf.compat.as_bytes(str(MODEL_VERSION)))
print('Exporting trained model to', export_path)
builder = tf.saved_model.builder.SavedModelBuilder(export_path)
tensor_info_x = tf.saved_model.utils.build_tensor_info(x_flat)
tensor_info_y = tf.saved_model.utils.build_tensor_info(model_train)
prediction_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs={'images': tensor_info_x},
outputs={'scores': tensor_info_y},
method_name=tf.saved_model.signature_constants.
PREDICT_METHOD_NAME))
legacy_init_op = tf.group(tf.tables_initializer(),
name='legacy_init_op')
builder.add_meta_graph_and_variables(
sess, [tf.saved_model.tag_constants.SERVING],
signature_def_map={
'predict_images': prediction_signature,
},
legacy_init_op=legacy_init_op)
builder.save()
print('Done exporting!')
print("Max batch accuracy is", max(acc_list))
print("Min batch accuracy is", min(acc_list))
print("Avg. accuracy:", sum(acc_list) / len(acc_list))