'drn':
transforms.Compose([
my_transform.Clahe(clipLimit=4, tileGridSize=(8, 8)),
my_transform.Resize(size=(512, 1024)),
]),
'seresnext':
transforms.Compose([
my_transform.Clahe(clipLimit=4, tileGridSize=(8, 8)),
])
}
image_datasets = {
'drn':
imagefolder.EdgeDataset(
input_dir=input_dir,
#label_dir=label_dir,
input_transform=data_transforms['drn'],
#label_transform=data_transforms[x],
mode='predict',
#randcrop=(1024,1024),
),
'seresnext':
imagefolder.EdgeDataset(
input_dir=input_dir,
#label_dir=label_dir,
input_transform=data_transforms['seresnext'],
#label_transform=data_transforms[x],
mode='predict',
#randcrop=(1024,1024),
)
}
dataset_size = len(image_datasets['drn'])
my_transform.Clahe(clipLimit=4, tileGridSize=(8, 8))
#transforms.Resize((32, 32),interpolation=Image.BILINEAR),
#transforms.CenterCrop(224),
#transforms.ToTensor()
#transforms.Normalize([0.5, 0.5], [0.5, 0.5])
]),
}
#data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in["train","validation"]}
image_datasets = {
x: imagefolder.EdgeDataset(
input_dir=input_dir,
label_dir=label_dir,
input_transform=data_transforms[x],
#label_transform=data_transforms[x],
mode=x,
randcrop=(1024, 1024) if x == 'train' else None,
n_class=4,
seed=seed,
randhflip=0.5 if x == 'train' else None,
per_train=per_train)
for x in ["train", "validation"]
}
dataset_sizes = {
x: len(image_datasets[x])
for x in ['train', 'validation']
}
print(dataset_sizes)
num_samples = {
'train':
#num_sample=2242
data_transform = transforms.Compose([
my_transform.Clahe(clipLimit=4, tileGridSize=(8, 8))
#transforms.Resize((32,32),interpolation=Image.BILINEAR),
#transforms.RandomVerticalFlip(),
#transforms.RandomHorizontalFlip(),
#transforms.ToTensor()
#transforms.Normalize([0.5, 0.5], [0.5, 0.5])
#transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
#data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in["train","validation"]}
image_dataset = imagefolder.EdgeDataset(
input_dir=input_dir,
#label_dir=label_dir,
input_transform=data_transform,
#label_transform=data_transforms[x],
mode='predict',
#randcrop=(1024,1024),
)
dataset_size = len(image_dataset)
print(dataset_size)
#num_samples={x:dataset_sizes[x]//20+batch_size-(dataset_sizes[x]//20)%batch_size for x in ['train', 'validation']}
#sample_weight={x:[1]*dataset_sizes[x] for x in ['train', 'validation']}
#data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in ["train","validation"]}
dataloader = torch.utils.data.DataLoader(image_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=2)
"validation":
transforms.Compose([
my_transform.Clahe(clipLimit=4, tileGridSize=(8, 8))
#transforms.Resize((32, 32),interpolation=Image.BILINEAR),
#transforms.CenterCrop(224),
#transforms.ToTensor()
#transforms.Normalize([0.5, 0.5], [0.5, 0.5])
]),
}
#data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in["train","validation"]}
image_datasets = {
x: imagefolder.EdgeDataset(
input_dir=input_dir,
label_dir=label_dir,
input_transform=data_transforms[x],
#label_transform=data_transforms[x],
mode=x,
per_train=per_train)
for x in ["train", "validation"]
}
dataset_sizes = {
x: len(image_datasets[x])
for x in ['train', 'validation']
}
print(dataset_sizes)
num_samples = {x: dataset_sizes[x] // 20 for x in ['train', 'validation']}
sample_weight = {
x: [1] * dataset_sizes[x]
for x in ['train', 'validation']
target_transforms={
'train': transforms.Compose([
my_transform.Resize(size=(512,1024)),
]),
"validation": transforms.Compose([
my_transform.Resize(size=(512,1024)),
]),
}
#data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in["train","validation"]}
image_datasets={x:imagefolder.EdgeDataset(input_dir=input_dir,
label_dir=label_dir,
input_transform=data_transforms[x],
label_transform=target_transforms[x] if parsed.use_resize else None,
mode=x,
randcrop=(1024,1024) if x=='train' and not parsed.use_resize else None,
n_class=4,
seed=seed,
randhflip=0.5 if x=='train' else None,
per_train=per_train)
for x in ["train","validation"]}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'validation']}
print(dataset_sizes)
num_samples={'train':dataset_sizes['train']//2+batch_size-(dataset_sizes['train']//2)%batch_size,
'validation':dataset_sizes['validation']//2+batch_size-(dataset_sizes['validation']//2)%batch_size}
# num_samples={'train':dataset_sizes['train'],
# 'validation':dataset_sizes['validation']}
sample_weight={x:[1]*dataset_sizes[x] for x in ['train', 'validation']}
data_sampler={x:torch.utils.data.sampler.WeightedRandomSampler(sample_weight[x], num_samples=num_samples[x], replacement=True) for x in ["train","validation"]}
