val_loss = 0
i = 0
for x_batch, y_batch in batch_eval:
batch_len = len(x_batch)
outputs = cnn(x_batch)
loss = loss_fn(outputs, y_batch)
val_loss += loss.item()
i += 1
_, prediction = torch.max(outputs.data, 1)
correct = (prediction == y_batch).sum().item()
acc = correct / batch_len
total_acc += acc * batch_len
return (total_acc / data_len), (val_loss / i)
cnn = resnet18().to(device)
#实验改动位置
#################################################################################
'''
针对optimizer进行实验
'''
optimizer = Adam(cnn.parameters(), lr=0.01) # 选用AdamOptimizer
#学习率优化若10轮训练loss未下降，则学习率*0.1
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer,
mode='min',
factor=0.1,
patience=100,
verbose=True,
import mxnet as mx
import mxnet.ndarray as nd
from mxnet.gluon import data as gdata
import cv2
from net import resnet18
img = cv2.imread('test_imgs/2.jpg', cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (48, 48))
print(img.shape)
img = nd.array(img).expand_dims(axis=2).expand_dims(axis=0)
print(img.shape)
transform_test = gdata.vision.transforms.Compose([
gdata.vision.transforms.ToTensor(),
gdata.vision.transforms.Normalize([0.5],
[0.5])])
img = transform_test(img)
net = resnet18(7)
ctx = mx.gpu(0)
net.load_parameters('trained_models/resnet18-epoch10-loss0.008498244381253217.params')
pred = net(img)[0]
idx = nd.argmax(pred, axis=0)
print(nd.argmax(pred, axis=0))
prob = out[0][pred].asscalar()
return '置信度=%f, 类别 %s' % (prob, label[str(pred)])
def predict_cv(net, ctx, fname, label):
img = cv2.imread(fname)
img = cv2.resize(img, (image_size, image_size))
data, _ = transform(nd.array(img), -1)
plt.imshow(data.transpose((1, 2, 0)).asnumpy() / 255)
data = data.expand_dims(axis=0)
out = net(data.as_in_context(ctx))
out = nd.SoftmaxActivation(out)
pred = int(nd.argmax(out, axis=1).asscalar())
prob = out[0][pred].asscalar()
print(prob, pred)
return '置信度=%f, 类别 %s' % (prob, label[str(pred)])
if __name__ == '__main__':
label_path = '/data/datasets/cifar-10/label.txt'
image_path = '/data/datasets/cifar-10/test/9/9_11.jpg'
label_dict = get_label(label_path)
print(label_dict)
ctx = utils.try_gpu()
net2 = net_collection.resnet18(10)
net2.hybridize()
net2.load_params('models/11_0.87632_0.89242.params', ctx=ctx)
print(predict_mxnet(net=net2, ctx=ctx, fname=image_path, label=label_dict))
plt.imshow(plt.imread(image_path))
plt.show()
help='Number of iterations we want to do for GTG')
args = parser.parse_args()
batch_size = 32
# torch.cuda.set_device(args.gpu_id)
if args.net_type == 'bn_inception':
model = net.bn_inception(pretrained=True, nb_classes=args.nb_classes)
model.last_linear = nn.Linear(1024, args.nb_classes)
if args.embed:
model = net.Inception_embed(model, 1024, args.sz_embedding,
args.nb_classes)
elif args.net_type == 'resnet18':
model = net.resnet18(pretrained=True)
model.fc = nn.Linear(512, args.nb_classes)
elif args.net_type == 'resnet34':
model = net.resnet34(pretrained=True)
model.fc = nn.Linear(512, args.nb_classes)
elif args.net_type == 'resnet50':
model = net.resnet50(pretrained=True)
model.fc = nn.Linear(2048, args.nb_classes)
elif args.net_type == 'resnet101':
model = net.resnet101(pretrained=True)
model.fc = nn.Linear(2048, args.nb_classes)
elif args.net_type == 'resnet152':
model = net.resnet152(pretrained=True)
model.fc = nn.Linear(2048, args.nb_classes)
elif args.net_type == 'densenet121':
model = net.densenet121(pretrained=True)
train_iter = gdata.DataLoader(train_ds.transform_first(transform_train),
batch_size,
shuffle=True,
last_batch='keep')
valid_iter = gdata.DataLoader(valid_ds.transform_first(transform_test),
batch_size,
shuffle=True,
last_batch='keep')
test_iter = gdata.DataLoader(test_ds.transform_first(transform_test),
batch_size,
shuffle=False,
last_batch='keep')
# 2. define training
ctx = mx.gpu(0)
net = resnet18(num_classes=7)
net.initialize(ctx=ctx, init=init.Xavier())
net.hybridize() # dynamic -> static
num_epochs = 100
init_lr = 0.01
wd = 5e-4
lr_period = 10
lr_decay = 0.5
load_params = True
trainer = gluon.Trainer(net.collect_params(), 'sgd', {
'learning_rate': init_lr,
'momentum': 0.9,
'wd': wd