ElasticTransform in PyTorch

#python #pytorch #elastictransform #v2

Buy Me a Coffee☕

*Memos:

My post explains OxfordIIITPet().

ElasticTransform() can do random morphological transformation for an image as shown below:

*Memos:

The 1st argument for initialization is alpha(Optional-Default:50.0-Type:int, float or tuple/list(int or float)): *Memos:
- It can do morphological transformation.
- It's the magnitude of displacements [number, number].
- It must be 0 <= number.
- A tuple/list must be the 1D with 1 or 2 elements.
- A single value(int, float or tuple/list(int or float)) means [alpha, alpha].
The 2nd argument for initialization is sigma(Optional-Default:0.5-Type:int, float or tuple/list(int or float)): *Memos:
- It's the smoothness of displacements [number, number].
- It must be 0 < number.
- A tuple/list must be the 1D with 1 or 2 elements.
- A single value(int, float or tuple/list(int or float)) means [sigma, sigma].
The 3rd argument for initialization is interpolation(Optional-Default:InterpolationMode.BILINEAR-Type:InterpolationMode).
The 4th 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 morphological transformation for an image.
- A tuple/list must be the 1D with 1 or 3 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 ElasticTransform
from torchvision.transforms.functional import InterpolationMode

elastictf = ElasticTransform()
elastictf = ElasticTransform(alpha=50,
                             sigma=5,
                             interpolation=InterpolationMode.BILINEAR,
                             fill=0)
elastictf
# ElasticTransform(alpha=[50.0, 50.0],
#                  sigma=[5.0, 5.0],
#                  interpolation=InterpolationMode.BILINEAR,
#                  fill=0)

elastictf.alpha
# [50.0, 50.0]

elastictf.sigma
# [5.0, 5.0]

elastictf.interpolation
# <InterpolationMode.BILINEAR: 'bilinear'>

elastictf.fill
# 0

origin_data = OxfordIIITPet(
    root="data",
    transform=None
    # transform=ElasticTransform(alpha=0, sigma=0)
)

a50_data = OxfordIIITPet( # `a` is alpha.
    root="data",
    transform=ElasticTransform(alpha=50)
    # transform=ElasticTransform(alpha=[50, 50])
)

a100_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100)
)

a200_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=200)
)

a500_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=500)
)

a1000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000)
)

a5000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000)
)

a10000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000)
)

a50000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=50000)
)

a100000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100000)
)

a1000000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000000)
)

a10000000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000000)
)

a50_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[50, 0])
)

a100_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[100, 0])
)

a200_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[200, 0])
)

a500_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[500, 0])
)

a1000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[1000, 0])
)

a5000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[5000, 0])
)

a10000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[10000, 0])
)

a50000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[50000, 0])
)

a100000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[100000, 0])
)

a1000000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[1000000, 0])
)

a10000000_0_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[10000000, 0])
)

a0_50_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 50])
)

a0_100_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 100])
)

a0_200_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 200])
)

a0_500_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 500])
)

a0_1000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 1000])
)

a0_5000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 5000])
)

a0_10000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 10000])
)

a0_50000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 50000])
)

a0_100000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 100000])
)

a0_1000000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 1000000])
)

a0_10000000_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=[0, 10000000])
)

a5000s01_data = OxfordIIITPet( # `s` is sigma.
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=0.1)
    # transform=ElasticTransform(alpha=5000, sigma=[0.1, 0.1])
)

a5000s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=1)
)

a5000s5_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=5)
)

a5000s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=10)
)

a5000s20_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=20)
)

a5000s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=40)
)

a5000s40_01_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 0.1])
)

a5000s40_1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 1])
)

a5000s40_5_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 5])
)

a5000s40_10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 10])
)

a5000s40_20_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 20])
)

a5000s40_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 40])
)

a5000s01_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[0.1, 40])
)

a5000s1_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[1, 40])
)

a5000s5_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[5, 40])
)

a5000s10_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[10, 40])
)

a5000s20_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[20, 40])
)

a5000s40_40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, sigma=[40, 40])
)

a5000fgray_data = OxfordIIITPet( # `f` is fill.
    root="data",
    transform=ElasticTransform(alpha=5000, fill=150)
)

a10000fgray_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, fill=150)
)

a5000fpurple_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=5000, fill=[160, 32, 240])
)

a10000fpurple_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, fill=[160, 32, 240])
)

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")
show_images1(data=a50_data, main_title="a50_data")
show_images1(data=a100_data, main_title="a100_data")
show_images1(data=a200_data, main_title="a200_data")
show_images1(data=a500_data, main_title="a500_data")
show_images1(data=a1000_data, main_title="a1000_data")
show_images1(data=a5000_data, main_title="a5000_data")
show_images1(data=a10000_data, main_title="a10000_data")
show_images1(data=a50000_data, main_title="a50000_data")
show_images1(data=a100000_data, main_title="a100000_data")
show_images1(data=a1000000_data, main_title="a1000000_data")
show_images1(data=a10000000_data, main_title="a10000000_data")
print()
show_images1(data=origin_data, main_title="origin_data")
show_images1(data=a50_0_data, main_title="a50_0_data")
show_images1(data=a100_0_data, main_title="a100_0_data")
show_images1(data=a200_0_data, main_title="a200_0_data")
show_images1(data=a500_0_data, main_title="a500_0_data")
show_images1(data=a1000_0_data, main_title="a1000_0_data")
show_images1(data=a5000_0_data, main_title="a5000_0_data")
show_images1(data=a10000_0_data, main_title="a10000_0_data")
show_images1(data=a50000_0_data, main_title="a50000_0_data")
show_images1(data=a100000_0_data, main_title="a100000_0_data")
show_images1(data=a1000000_0_data, main_title="a1000000_0_data")
show_images1(data=a10000000_0_data, main_title="a10000000_0_data")
print()
show_images1(data=origin_data, main_title="origin_data")
show_images1(data=a0_50_data, main_title="a0_50_data")
show_images1(data=a0_100_data, main_title="a0_100_data")
show_images1(data=a0_200_data, main_title="a0_200_data")
show_images1(data=a0_500_data, main_title="a0_500_data")
show_images1(data=a0_1000_data, main_title="a0_1000_data")
show_images1(data=a0_5000_data, main_title="a0_5000_data")
show_images1(data=a0_10000_data, main_title="a0_10000_data")
show_images1(data=a0_50000_data, main_title="a0_50000_data")
show_images1(data=a0_100000_data, main_title="a0_100000_data")
show_images1(data=a0_1000000_data, main_title="a0_1000000_data")
show_images1(data=a0_10000000_data, main_title="a0_10000000_data")
print()
show_images1(data=a5000s01_data, main_title="a5000s01_data")
show_images1(data=a5000s1_data, main_title="a5000s1_data")
show_images1(data=a5000s5_data, main_title="a5000s5_data")
show_images1(data=a5000s10_data, main_title="a5000s10_data")
show_images1(data=a5000s20_data, main_title="a5000s20_data")
show_images1(data=a5000s40_data, main_title="a5000s40_data")
print()
show_images1(data=a5000s40_01_data, main_title="a5000s40_01_data")
show_images1(data=a5000s40_1_data, main_title="a5000s40_1_data")
show_images1(data=a5000s40_5_data, main_title="a5000s40_5_data")
show_images1(data=a5000s40_10_data, main_title="a5000s40_10_data")
show_images1(data=a5000s40_20_data, main_title="a5000s40_20_data")
show_images1(data=a5000s40_40_data, main_title="a5000s40_40_data")
print()
show_images1(data=a5000s01_40_data, main_title="a5000s01_40_data")
show_images1(data=a5000s1_40_data, main_title="a5000s1_40_data")
show_images1(data=a5000s5_40_data, main_title="a5000s5_40_data")
show_images1(data=a5000s10_40_data, main_title="a5000s10_40_data")
show_images1(data=a5000s20_40_data, main_title="a5000s20_40_data")
show_images1(data=a5000s40_40_data, main_title="a5000s40_40_data")
print()
show_images1(data=a5000fgray_data, main_title="a5000fgray_data")
show_images1(data=a10000fgray_data, main_title="a10000fgray_data")
show_images1(data=a5000fpurple_data, main_title="a5000fpurple_data")
show_images1(data=a10000fpurple_data, main_title="a10000fpurple_data")

# ↓ ↓ ↓ ↓ ↓ ↓ The code below is identical to the code above. ↓ ↓ ↓ ↓ ↓ ↓
def show_images2(data, main_title=None, a=50, s=5, f=0):
    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)
        et = ElasticTransform(alpha=a, sigma=s, fill=f) # Here
        plt.imshow(X=et(im)) # Here
        plt.xticks(ticks=[])
        plt.yticks(ticks=[])
    plt.tight_layout()
    plt.show()

show_images2(data=origin_data, main_title="origin_data", a=0, s=0)
show_images2(data=origin_data, main_title="a50_data", a=50)
show_images2(data=origin_data, main_title="a100_data", a=100)
show_images2(data=origin_data, main_title="a200_data", a=200)
show_images2(data=origin_data, main_title="a500_data", a=500)
show_images2(data=origin_data, main_title="a1000_data", a=1000)
show_images2(data=origin_data, main_title="a5000_data", a=5000)
show_images2(data=origin_data, main_title="a10000_data", a=10000)
show_images2(data=origin_data, main_title="a50000_data", a=50000)
show_images2(data=origin_data, main_title="a100000_data", a=100000)
show_images2(data=origin_data, main_title="a1000000_data", a=1000000)
show_images2(data=origin_data, main_title="a10000000_data", a=10000000)
print()
show_images2(data=origin_data, main_title="origin_data", a=0, s=0)
show_images2(data=origin_data, main_title="a50_0_data", a=[50, 0])
show_images2(data=origin_data, main_title="a100_0_data", a=[100, 0])
show_images2(data=origin_data, main_title="a200_0_data", a=[200, 0])
show_images2(data=origin_data, main_title="a500_0_data", a=[500, 0])
show_images2(data=origin_data, main_title="a1000_0_data", a=[1000, 0])
show_images2(data=origin_data, main_title="a5000_0_data", a=[5000, 0])
show_images2(data=origin_data, main_title="a10000_0_data", a=[10000, 0])
show_images2(data=origin_data, main_title="a50000_0_data", a=[50000, 0])
show_images2(data=origin_data, main_title="a100000_0_data", a=[100000, 0])
show_images2(data=origin_data, main_title="a1000000_0_data", a=[1000000, 0])
show_images2(data=origin_data, main_title="a10000000_0_data", a=[10000000, 0])
print()
show_images2(data=origin_data, main_title="origin_data", a=0, s=0)
show_images2(data=origin_data, main_title="a0_50_data", a=[0, 50])
show_images2(data=origin_data, main_title="a0_100_data", a=[0, 100])
show_images2(data=origin_data, main_title="a0_200_data", a=[0, 200])
show_images2(data=origin_data, main_title="a0_500_data", a=[0, 500])
show_images2(data=origin_data, main_title="a0_1000_data", a=[0, 1000])
show_images2(data=origin_data, main_title="a0_5000_data", a=[0, 5000])
show_images2(data=origin_data, main_title="a0_10000_data", a=[0, 10000])
show_images2(data=origin_data, main_title="a0_50000_data", a=[0, 50000])
show_images2(data=origin_data, main_title="a0_100000_data", a=[0, 100000])
show_images2(data=origin_data, main_title="a0_1000000_data", a=[0, 1000000])
show_images2(data=origin_data, main_title="a0_10000000_data", a=[0, 10000000])
print()
show_images2(data=origin_data, main_title="a5000s01_data", a=5000, s=0.1)
show_images2(data=origin_data, main_title="a5000s1_data", a=5000, s=1)
show_images2(data=origin_data, main_title="a5000s5_data", a=5000, s=5)
show_images2(data=origin_data, main_title="a5000s10_data", a=5000, s=10)
show_images2(data=origin_data, main_title="a5000s20_data", a=5000, s=20)
show_images2(data=origin_data, main_title="a5000s40_data", a=5000, s=40)
print()
show_images2(data=origin_data, main_title="a5000s40_01_data", a=5000,
             s=[40, 0.1])
show_images2(data=origin_data, main_title="a5000s40_1_data", a=5000,
             s=[40, 1])
show_images2(data=origin_data, main_title="a5000s40_5_data", a=5000,
             s=[40, 5])
show_images2(data=origin_data, main_title="a5000s40_10_data", a=5000,
             s=[40, 10])
show_images2(data=origin_data, main_title="a5000s40_20_data", a=5000,
             s=[40, 20])
show_images2(data=origin_data, main_title="a5000s40_40_data", a=5000,
             s=[40, 40])
print()
show_images2(data=origin_data, main_title="a5000s01_40_data", a=5000,
             s=[0.1, 40])
show_images2(data=origin_data, main_title="a5000s1_40_data", a=5000,
             s=[1, 40])
show_images2(data=origin_data, main_title="a5000s5_40_data", a=5000,
             s=[5, 40])
show_images2(data=origin_data, main_title="a5000s10_40_data", a=5000,
             s=[10, 40])
show_images2(data=origin_data, main_title="a5000s20_40_data", a=5000,
             s=[20, 40])
show_images2(data=origin_data, main_title="a5000s40_40_data", a=5000,
             s=[40, 40])
print()
show_images2(data=origin_data, main_title="a5000fgray_data", a=5000, f=150)
show_images2(data=origin_data, main_title="a10000fgray_data", a=10000, f=150)
show_images2(data=origin_data, main_title="a5000fpurple_data", a=5000,
             f=[160, 32, 240])
show_images2(data=origin_data, main_title="a10000fpurple_data", a=10000,
             f=[160, 32, 240])