def train():
from paddlex.cls import transforms
train_transforms = transforms.Compose([
transforms.RandomCrop(crop_size=224),
transforms.RandomHorizontalFlip(),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.ResizeByShort(short_size=256),
transforms.CenterCrop(crop_size=224),
transforms.Normalize()
])
train_dataset = pdx.datasets.ImageNet(
data_dir=f'{data_dir}/mini_imagenet_veg',
file_list=f'{data_dir}/mini_imagenet_veg/train_list.txt',
label_list=f'{data_dir}/mini_imagenet_veg/labels.txt',
transforms=train_transforms)
eval_dataset = pdx.datasets.ImageNet(
data_dir=f'{data_dir}/mini_imagenet_veg',
file_list=f'{data_dir}/mini_imagenet_veg/val_list.txt',
label_list=f'{data_dir}/mini_imagenet_veg/labels.txt',
transforms=eval_transforms)
num_classes = len(train_dataset.labels)
model = pdx.cls.MobileNetV3_large_ssld(num_classes=num_classes)
model.train(num_epochs=12,
train_dataset=train_dataset,
train_batch_size=32,
eval_dataset=eval_dataset,
lr_decay_epochs=[6, 8],
save_interval_epochs=1,
learning_rate=0.00625,
save_dir=f'{output_dir}/output/mobilenetv3_large_ssld',
use_vdl=True)
def train():
from paddlex.cls import transforms
train_transforms = transforms.Compose([
transforms.RandomCrop(crop_size=224),
transforms.RandomHorizontalFlip(),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.ResizeByShort(short_size=256),
transforms.CenterCrop(crop_size=224),
transforms.Normalize()
])
train_dataset = pdx.datasets.ImageNet(
data_dir=f'{data_dir}/vegetables_cls',
file_list=f'{data_dir}/vegetables_cls/train_list.txt',
label_list=f'{data_dir}/vegetables_cls/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.ImageNet(
data_dir=f'{data_dir}/vegetables_cls',
file_list=f'{data_dir}/vegetables_cls/val_list.txt',
label_list=f'{data_dir}/vegetables_cls/labels.txt',
transforms=eval_transforms)
num_classes = len(train_dataset.labels)
model = pdx.cls.MobileNetV2(num_classes=num_classes)
model.train(num_epochs=10,
train_dataset=train_dataset,
train_batch_size=32,
eval_dataset=eval_dataset,
lr_decay_epochs=[4, 6, 8],
save_interval_epochs=1,
learning_rate=0.025,
save_dir=f'{output_dir}/mobilenetv2',
use_vdl=True)
def train(model_dir=None, sensitivities_file=None, eval_metric_loss=0.05):
# 下载和解压蔬菜分类数据集
veg_dataset = 'https://bj.bcebos.com/paddlex/datasets/vegetables_cls.tar.gz'
pdx.utils.download_and_decompress(veg_dataset, path='./')
# 定义训练和验证时的transforms
train_transforms = transforms.Compose([
transforms.RandomCrop(crop_size=224),
transforms.RandomHorizontalFlip(),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.ResizeByShort(short_size=256),
transforms.CenterCrop(crop_size=224),
transforms.Normalize()
])
# 定义训练和验证所用的数据集
train_dataset = pdx.datasets.ImageNet(
data_dir='vegetables_cls',
file_list='vegetables_cls/train_list.txt',
label_list='vegetables_cls/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.ImageNet(
data_dir='vegetables_cls',
file_list='vegetables_cls/val_list.txt',
label_list='vegetables_cls/labels.txt',
transforms=eval_transforms)
num_classes = len(train_dataset.labels)
model = pdx.cls.MobileNetV2(num_classes=num_classes)
if model_dir is None:
# 使用imagenet数据集预训练模型权重
pretrain_weights = "IMAGENET"
else:
# 使用传入的model_dir作为预训练模型权重
assert os.path.isdir(model_dir), "Path {} is not a directory".format(
model_dir)
pretrain_weights = model_dir
save_dir = './output/mobilenetv2'
if sensitivities_file is not None:
# DEFAULT 指使用模型预置的参数敏感度信息作为裁剪依据
if sensitivities_file != "DEFAULT":
assert os.path.exists(
sensitivities_file), "Path {} not exist".format(
sensitivities_file)
save_dir = './output/mobilenetv2_prune'
model.train(
num_epochs=10,
train_dataset=train_dataset,
train_batch_size=32,
eval_dataset=eval_dataset,
lr_decay_epochs=[4, 6, 8],
learning_rate=0.025,
pretrain_weights=pretrain_weights,
save_dir=save_dir,
use_vdl=True,
sensitivities_file=sensitivities_file,
eval_metric_loss=eval_metric_loss)
with open(os.path.join('train_list.txt'), 'w') as f:
for i, cls_fold in enumerate(os.listdir(base)):
cls_base = os.path.join(base, cls_fold)
files = os.listdir(cls_base)
print('{} train num:'.format(cls_fold), len(files))
for pt in files:
img = os.path.join(cls_fold, pt)
info = img + ' ' + str(i) + '\n'
f.write(info)
with open(os.path.join('labels.txt'), 'w') as f:
for i, cls_fold in enumerate(os.listdir(base)):
f.write(cls_fold + '\n')
train_transforms = transforms.Compose(
[transforms.RandomCrop(crop_size=224),
transforms.Normalize()])
train_dataset = pdx.datasets.ImageNet(data_dir=base,
file_list='train_list.txt',
label_list='labels.txt',
transforms=train_transforms,
shuffle=True)
num_classes = len(train_dataset.labels)
model = pdx.cls.ResNet18(num_classes=num_classes)
model.train(num_epochs=20,
train_dataset=train_dataset,
train_batch_size=32,
lr_decay_epochs=[5, 10, 15],
learning_rate=2e-2,
def loader_test(path):
x = Image.open(path).convert("RGB")
x = np.asarray(x).astype('float32')
#x = cv2.imread("work/"+path,cv2.IMREAD_COLOR)
x = cv2.cvtColor(x, cv2.COLOR_RGB2BGR) / 255.0
x = (cv2.resize(x, (1024, 1024)) - 0.5) / 0.5
return x
transform_ops = transforms.Compose([
#transforms.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5]) # this will do 1/255.0
transforms.RandomHorizontalFlip(prob=0.5),
transforms.RandomRotate(rotate_range=30, prob=0.5),
transforms.RandomCrop(crop_size=224,
lower_scale=0.7,
lower_ratio=3. / 4,
upper_ratio=4. / 3),
transforms.RandomDistort(brightness_range=0.1,
brightness_prob=0.5,
contrast_range=0.1,
contrast_prob=0.5,
saturation_range=0.1,
saturation_prob=0.0,
hue_range=0.1,
hue_prob=0.0)
])
def transform(img):
#print("before transform: ",img.shape)
img = transform_ops(img)[0]
