Programming

RandomAffine in PyTorch (1)

Buy Me a Coffee☕ Memos: My post explains RandomAffine() with scale argument. My post explains RandomRotation(). My post explains RandomPerspective(). My post explains OxfordIIITPet(). RandomAffine() can do random rotation or random affine transformation for an image as shown below: Memos: The 1st argument for initialization is degrees(Required-Type:int, float or tuple/list(int or float)): *Memos: It can do rotation. It's the range of the degrees [min, max] so it must min

Feb 7, 2025 - 15:27

Buy Me a Coffee☕

*Memos:

My post explains RandomAffine() with scale argument.
My post explains RandomRotation().
My post explains RandomPerspective().
My post explains OxfordIIITPet().

RandomAffine() can do random rotation or random affine transformation for an image as shown below:

*Memos:

The 1st argument for initialization is degrees(Required-Type:int, float or tuple/list(int or float)): *Memos:
- It can do rotation.
- It's the range of the degrees [min, max] so it must min <= max.
- A degrees value is randomly taken from the range of [min, max].
- A tuple/list must be the 1D with 2 elements.
- A single value(int or float) means [-degrees(min), +degrees(max)].
- A single value(int or float) must be 0 <= x.
The 2nd argument for initialization is translate(Optional-Default:None-Type:tuple/list(int or float)): *Memos:
- It's [a, b].
- It must be 0 <= 1.
- It must be the 1D with 2 elements.
- The 1st element is for the horizontal shift randomly taken in the range of -img_width * a < horizontal shift < img_width * a.
- The 2nd element is for the vertical shift randomly taken in the range of -img_height * b < vertical shift < img_height * b.
The 3rd argument for initialization is scale(Optional-Default:None-Type:tuple/list(int or float)): *Memos:
- It's [min, max] so it must min <= max.
- It must be 0 < x.
- It must be the 1D with 2 elements.
- A scale value is randomly taken from the range of [min, max].
The 4th argument for initialization is shear(Optional-Default:None-Type:int, float or tuple/list(int or float)): *Memos:
- It can do affine transformation with x and y.
- It's [min, max, min, max] so it must min <= max. *Memos:
- The 1st two elements are the range of x.
- The 2nd two elements are the range of y.
- x value is randomly taken from the range of the 1st two elements.
- y value is randomly taken from the range of the 2nd two elements.
- A tuple/list must be the 1D with 2 or 4 elements.
- The tuple/list of 2 elements means [shear[0](min), shear[1](max), 0.0(min), 0.0(min)].
- A single value means [-shear(min), +shear(max), 0.0(min), 0.0(max)].
- A single value must be 0 <= x.
The 5th argument for initialization is interpolation(Optional-Default:InterpolationMode.NEAREST-Type:InterpolationMode).
The 6th argument for initialization is fill(Optional-Default:0-Type:int, float or tuple/list(int or float)): *Memos:
- It can change the background of an image. *The background can be seen when doing rotation or affine transformation for an image.
- A tuple/list must be the 1D with 1 or 3 elements.
The 7th argument for initialization is center(Optional-Default:None-Type:tuple/list(int or float)): *Memos:
- It can change the center position of an image.
- It must be the 1D with 2 elements.
The 1st argument is img(Required-Type:PIL Image or tensor(int)): *Memos:
- A tensor must be 3D.
- Don't use img=.
v2 is recommended to use according to V1 or V2? Which one should I use?.

from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import RandomAffine
from torchvision.transforms.functional import InterpolationMode

randomaffine = RandomAffine(degrees=90)
randomaffine = RandomAffine(degrees=[-90, 90],
                            translate=None,
                            scale=None,
                            shear=None,
                            interpolation=InterpolationMode.NEAREST,
                            fill=0,
                            center=None)
randomaffine
# RandomAffine(degrees=[-90, 90],
#              interpolation=InterpolationMode.NEAREST,
#              fill=0)

randomaffine.degrees
# [-90.0, 90.0]

print(randomaffine.translate)
# None

print(randomaffine.scale)
# None

print(randomaffine.shear)
# None

randomaffine.interpolation
# 

randomaffine.fill
# 0

print(randomaffine.center)
# None

origin_data = OxfordIIITPet(
    root="data",
    transform=None
    # transform=RandomAffine(degrees=[0, 0])
)

d90_data = OxfordIIITPet( # `d` is degrees.
    root="data",
    transform=RandomAffine(degrees=90)
    # transform=RandomAffine(degrees=[-90, 90])
)

dn90_0_data = OxfordIIITPet( # `n` is negative.
    root="data",
    transform=RandomAffine(degrees=[-90, 0])
)

d0_90_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[0, 90])
)

dn90n90_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[-90, -90])
)

d90_90_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[90, 90])
)

d180_180_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[180, 180])
)

hrvrtran_data = OxfordIIITPet( # `hr` is horizontal and `vr` is vertical.
    root="data",               # `tran` is translate.
    transform=RandomAffine(degrees=[0, 0], translate=[0.8, 0.5])
)

hrtran_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[0, 0], translate=[0.8, 0])
)

vrtran_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[0, 0], translate=[0, 0.5])
)

dn45n45fgray_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[-45, -45], fill=150)
)

d135_135fpurple_data = OxfordIIITPet(
    root="data",
    transform=RandomAffine(degrees=[135, 135], fill=[160, 32, 240])
)

d180_180c270_200_data = OxfordIIITPet( # `c` is center.
    root="data",
    transform=RandomAffine(degrees=[180, 180], center=[270, 200])
)

import matplotlib.pyplot as plt

def show_images1(data, main_title=None):
    plt.figure(figsize=[10, 5])
    plt.suptitle(t=main_title, y=0.8, fontsize=14)
    for i, (im, _) in zip(range(1, 6), data):
        plt.subplot(1, 5, i)
        plt.imshow(X=im)
        plt.xticks(ticks=[])
        plt.yticks(ticks=[])
    plt.tight_layout()
    plt.show()

show_images1(data=origin_data, main_title="origin_data")
print()
show_images1(data=d90_data, main_title="d90_data")
show_images1(data=dn90_0_data, main_title="dn90_0_data")
show_images1(data=d0_90_data, main_title="d0_90_data")
print()
show_images1(data=dn90n90_data, main_title="dn90n90_data")
show_images1(data=d90_90_data, main_title="d90_90_data")
show_images1(data=d180_180_data, main_title="d180_180_data")
print()
show_images1(data=hrvrtran_data, main_title="hrvrtran_data")
show_images1(data=hrtran_data, main_title="hrtran_data")
show_images1(data=vrtran_data, main_title="vrtran_data")
print()
show_images1(data=dn45n45fgray_data, main_title="dn45n45fgray_data")
show_images1(data=d135_135fpurple_data, main_title="d135_135fpurple_data")
show_images1(data=d180_180c270_200_data, main_title="d180_180c270_200_data")

# ↓ ↓ ↓ ↓ ↓ ↓ The code below is identical to the code above. ↓ ↓ ↓ ↓ ↓ ↓ 
def show_images2(data, main_title=None, d=0, t=None, sc=None, sh=None,
                 ip=InterpolationMode.NEAREST, f=0, c=None):
    plt.figure(figsize=[10, 5])
    plt.suptitle(t=main_title, y=0.8, fontsize=14)
    for i, (im, _) in zip(range(1, 6), data):
        plt.subplot(1, 5, i)
        ra = RandomAffine(degrees=d, translate=t, scale=sc, # Here
                          shear=sh, interpolation=ip, center=c, fill=f)
        plt.imshow(X=ra(im)) # Here
        plt.xticks(ticks=[])
        plt.yticks(ticks=[])
    plt.tight_layout()
    plt.show()

show_images2(data=origin_data, main_title="origin_data")
print()
show_images2(data=origin_data, main_title="d90_data", d=90)
show_images2(data=origin_data, main_title="dn90_0_data", d=[-90, 0])
show_images2(data=origin_data, main_title="d0_90_data", d=[0, 90])
print()
show_images2(data=origin_data, main_title="dn90n90_data", d=[-90, -90])
show_images2(data=origin_data, main_title="d90_90_data", d=[90, 90])
show_images2(data=origin_data, main_title="d180_180_data", d=[180, 180])
print()
show_images2(data=origin_data, main_title="hrvrtran_data", d=[0, 0],
             t=[0.8, 0.5])
show_images2(data=origin_data, main_title="hrtran_data", d=[0, 0],
             t=[0.8, 0])
show_images2(data=origin_data, main_title="vrtran_data", d=[0, 0],
             t=[0, 0.5])
print()
show_images2(data=origin_data, main_title="dn45n45fgray_data",
             d=[-45, -45], f=150)
show_images2(data=origin_data, main_title="d135_135fpurple_data",
             d=[135, 135], f=[160, 32, 240])
show_images2(data=origin_data, main_title="d180_180c270_200_data",
             d=[180, 180], c=[270, 200])

RandomAffine in PyTorch (1)

Tags:

Related Posts

Popular Posts

Recommended Posts