*Memos:
- My post explains ToPILImage() about no arguments.
- My post explains PILToTensor().
- My post explains device conversion with to(), from_numpy() and numpy().
- My post explains permute().
- My post explains OxfordIIITPet().
You can do conversion with PIL(Pillow library) Image, PyTorch tensor and NumPy array as shown below:
from torchvision.datasets import OxfordIIITPet
origin_data = OxfordIIITPet(
root="data",
transform=None
)
import matplotlib.pyplot as plt
plt.figure(figsize=[7, 9])
plt.title(label="s500_394origin_data", fontsize=14)
plt.imshow(X=origin_data[0][0])
plt.show()
PIL image[H, W, C] => PyTorch tensor[C, H, W] => NumPy array[H, W, C] => PIL image[H, W, C]:
from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import PILToTensor, ToPILImage
import numpy as np
from PIL import Image
origin_data = OxfordIIITPet(
root="data",
transform=None
)
# PIL image to PyTorch tensor
ptt = PILToTensor()
pytorchimagetensor = ptt(origin_data[0][0])
# tensor([[[ 37, 35, 36, ..., 247, 249, 249],
# [ 35, 35, 37, ..., 246, 248, 249],
# ...,
# [ 28, 28, 27, ..., 59, 65, 76]],
# [[ 20, 18, 19, ..., 248, 248, 248],
# [ 18, 18, 20, ..., 247, 247, 248],
# ...,
# [ 27, 27, 27, ..., 94, 106, 117]],
# [[ 12, 10, 11, ..., 253, 253, 253],
# [ 10, 10, 12, ..., 251, 252, 253],
# ...,
# [ 35, 35, 35, ..., 214, 232, 223]]], dtype=torch.uint8)
# PyTorch tensor to NumPy array
numpyimagearray = pytorchimagetensor.permute(1, 2, 0).numpy()
numpyimagearray = np.array(object=pytorchimagetensor.permute(dims=[1, 2, 0)])
numpyimagearray = np.asarray(pytorchimagetensor.permute(dims=[1, 2, 0]))
numpyimagearray
# array([[[ 37 20 12]
# [ 35 18 10]
# ...
# [249 248 253]]
# [[ 35 18 10]
# [ 35 18 10]
# ...
# [249 248 253]]
# [[ 35 18 10]
# [ 36 19 11]
# ...
# [250 249 254]]
# ...
# [[ 5 6 24]
# [ 4 5 23]
# ...
# [ 69 110 224]]
# [[ 4 3 19]
# [ 3 2 18]
# ...
# [ 64 108 229]]
# [[ 28 27 35]
# [ 28 27 35]
# ...
# [ 76 117 223]]], dtype=uint8)
# NumPy array to PIL image
tpi = ToPILImage()
pilimage = tpi(numpyimagearray)
pilimage = Image.fromarray(obj=numpyimagearray)
import matplotlib.pyplot as plt
plt.figure(figsize=[7, 9])
plt.title(label="s500_394origin_data", fontsize=14)
plt.imshow(X=pilimage)
plt.show()
PIL image[H, W, C] => NumPy array[H, W, C] => PyTorch tensor[C, H, W] => PIL image[H, W, C]:
from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import ToPILImage
import numpy as np
import torch
origin_data = OxfordIIITPet(
root="data",
transform=None
)
# PIL image to NumPy array
numpyimagearray = np.array(object=origin_data[0][0])
numpyimagearray = np.asarray(origin_data[0][0])
numpyimagearray
# array([[[ 37 20 12]
# [ 35 18 10]
# ...
# [249 248 253]]
# [[ 35 18 10]
# [ 35 18 10]
# ...
# [249 248 253]]
# [[ 35 18 10]
# [ 36 19 11]
# ...
# [250 249 254]]
# ...
# [[ 5 6 24]
# [ 4 5 23]
# ...
# [ 69 110 224]]
# [[ 4 3 19]
# [ 3 2 18]
# ...
# [ 64 108 229]]
# [[ 28 27 35]
# [ 28 27 35]
# ...
# [ 76 117 223]]], dtype=uint8)
# NumPy array to PyTorch tensor
pytorchimagetensor = torch.from_numpy(numpyimagearray).permute(dims=[2, 0, 1])
pytorchimagetensor = torch.tensor(numpyimagearray).permute(dims=[2, 0, 1])
pytorchimagetensor
# tensor([[[ 37, 35, 36, ..., 247, 249, 249],
# [ 35, 35, 37, ..., 246, 248, 249],
# ...,
# [ 28, 28, 27, ..., 59, 65, 76]],
# [[ 20, 18, 19, ..., 248, 248, 248],
# [ 18, 18, 20, ..., 247, 247, 248],
# ...,
# [ 27, 27, 27, ..., 94, 106, 117]],
# [[ 12, 10, 11, ..., 253, 253, 253],
# [ 10, 10, 12, ..., 251, 252, 253],
# ...,
# [ 35, 35, 35, ..., 214, 232, 223]]], dtype=torch.uint8)
# PyTorch tensor to PIL image
tpi = ToPILImage()
pilimage = tpi(pytorchimagetensor)
import matplotlib.pyplot as plt
plt.figure(figsize=[7, 9])
plt.title(label="s500_394origin_data", fontsize=14)
plt.imshow(X=pilimage)
plt.show()
Top comments (0)