test_custom_transform.py

import torchvision.datasets as datasets
import torchvision.transforms as transforms
import torch
import torchvision
import matplotlib.pyplot as plt
import numpy as np

from custom_transforms import NRandomCrop


def imshow(img):
    img = img / 2 + 0.5
    npimg = img.numpy()
    plt.imshow(np.transpose(npimg, (1, 2, 0)))
    plt.show()


mean, sd = (0.5, 0.5, 0.5), (0.5, 0.5, 0.5)
transform = transforms.Compose(
    [NRandomCrop(size=32, n=5, padding=4),
     transforms.Lambda(
         lambda crops: torch.stack([transforms.Normalize(mean, sd)(transforms.ToTensor()(crop)) for crop in crops])),
     ]
)


train_data = datasets.CIFAR10(root='./data',
                                  train=True,
                                  download=True,
                                  transform=transform)
train_loader = torch.utils.data.DataLoader(train_data,
                                           batch_size=1,
                                           shuffle=True,
                                           num_workers=4)
classes = ('plane', 'car', 'bird', 'cat',
               'deer', 'dog', 'frog', 'horse', 'ship', 'truck')


dataiter = iter(train_loader)
images, labels = dataiter.next()

# show images
imshow(torchvision.utils.make_grid(images.squeeze(0)))
# print labels
print(' '.join('%5s' % classes[labels[j]] for j in range(0)))

May I ask how to resize crops after NRandomCrop?

May I ask how to resize crops after NRandomCrop?

You can use a transforms.Resize(size) when stacking them after transforms.Normalize() in line 22.

'trainval': transforms.Compose([ NRandomCrop(size=args.crop_size, n=args.num_crop), transforms.Lambda( lambda crops: torch.stack([transforms.ToPILImage() (transforms.Resize((args.scale_size, args.scale_size))) (transforms.ToTensor()) (transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])(crop)) for crop in crops] ) ), ]),
This is my code but I donot know how to solve this error:
TypeError: pic should be Tensor or ndarray. Got <class 'torchvision.transforms.transforms.Resize'>.

            'trainval': transforms.Compose([
                NRandomCrop(size=args.crop_size, n=args.num_crop),
                transforms.Lambda(
                    lambda crops: torch.stack([transforms.ToPILImage()
                                               (transforms.Resize((args.scale_size, args.scale_size)))
                                               (transforms.ToTensor())
                                               (transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])(crop))
                                               for crop in crops]
                                              )
                ),
            ]),

Resize is done first, then ToTensor and then Normalize:

transform = transforms.Compose(
    [NRandomCrop(size=32, n=5, padding=4),
     transforms.Lambda(
         lambda crops: torch.stack([transforms.Normalize(mean, sd)(transforms.ToTensor()(transforms.Resize(224, 224)(crop))) for crop in crops])),
     ]
)

I get a 5-dimensional tensor but expected a 4-dimensional tensor. May be I need cat n crops to one tensor?

I get a 5-dimensional tensor but expected a 4-dimensional tensor. May be I need cat n crops to one tensor?

You can do something this in you training code. It only works with batch_size=1.

for batch in loader:
            images = batch['images']
            if len(images.shape) == 5: 
                images = torch.squeeze(images)
                assert len(images.shape)<5, "Please use batch size 1"