ElasticTransform in PyTorch (4)

#python #pytorch #elastictransform #v2

*Memos:

My post explains ElasticTransform() about no arguments, alpha argument with sigma=0.1 and sigma argument with alpha=0.
My post explains ElasticTransform() about alpha and fill argument.
My post explains ElasticTransform() about sigma and fill argument.

ElasticTransform() can do random morphological transformation for an image as shown below. *It's about alpha and sigma argument:

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

origin_data = OxfordIIITPet(
    root="data",
    transform=None
)

a0s01_data = OxfordIIITPet( # `a` is alpha and `s` is sigma.
    root="data",
    transform=ElasticTransform(alpha=0, sigma=0.1)
    # transform=ElasticTransform(alpha=[0, 0], sigma=[0.1, 0.1])
)

a0s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=0, sigma=1)
)

a0s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=0, sigma=10)
)

a0s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=0, sigma=40)
)

a10s01_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10, sigma=0.1)
    # transform=ElasticTransform(alpha=-10, sigma=0.1)
)

a10s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10, sigma=1)
)

a10s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10, sigma=10)
)

a10s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10, sigma=40)
)

a100s01_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100, sigma=0.1)
    # transform=ElasticTransform(alpha=-100, sigma=0.1)
)

a100s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100, sigma=1)
)

a100s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100, sigma=10)
)

a100s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=100, sigma=40)
)

a1000s01_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000, sigma=0.1)
    # transform=ElasticTransform(alpha=-1000, sigma=0.1)
)

a1000s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000, sigma=1)
)

a1000s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000, sigma=10)
)

a1000s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=1000, sigma=40)
)

a10000s01_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, sigma=0.1)
    # transform=ElasticTransform(alpha=-10000, sigma=0.1)
)

a10000s1_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, sigma=1)
)

a10000s10_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, sigma=10)
)

a10000s40_data = OxfordIIITPet(
    root="data",
    transform=ElasticTransform(alpha=10000, sigma=40)
)

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=a0s01_data, main_title="a0s01_data")
show_images1(data=a0s1_data, main_title="a0s1_data")
show_images1(data=a0s10_data, main_title="a0s10_data")
show_images1(data=a0s40_data, main_title="a0s40_data")
print()
show_images1(data=a10s01_data, main_title="a10s01_data")
show_images1(data=a10s1_data, main_title="a10s1_data")
show_images1(data=a10s10_data, main_title="a10s10_data")
show_images1(data=a10s40_data, main_title="a10s40_data")
print()
show_images1(data=a100s01_data, main_title="a100s01_data")
show_images1(data=a100s1_data, main_title="a100s1_data")
show_images1(data=a100s10_data, main_title="a100s10_data")
show_images1(data=a100s40_data, main_title="a100s40_data")
print()
show_images1(data=a1000s01_data, main_title="a1000s01_data")
show_images1(data=a1000s1_data, main_title="a1000s1_data")
show_images1(data=a1000s10_data, main_title="a1000s10_data")
show_images1(data=a1000s40_data, main_title="a1000s40_data")
print()
show_images1(data=a10000s01_data, main_title="a10000s01_data")
show_images1(data=a10000s1_data, main_title="a10000s1_data")
show_images1(data=a10000s10_data, main_title="a10000s10_data")
show_images1(data=a10000s40_data, main_title="a10000s40_data")

# ↓ ↓ ↓ ↓ ↓ ↓ The code below is identical to the code above. ↓ ↓ ↓ ↓ ↓ ↓
def show_images2(data, main_title=None, a=50, s=5, 
                 ip=InterpolationMode.BILINEAR, f=0):
    plt.figure(figsize=[10, 5])
    plt.suptitle(t=main_title, y=0.8, fontsize=14)
    if main_title != "origin_data":
        for i, (im, _) in zip(range(1, 6), data):
            plt.subplot(1, 5, i)
            et = ElasticTransform(alpha=a, sigma=s,
                                  interpolation=ip, fill=f)
            plt.imshow(X=et(im))
            plt.xticks(ticks=[])
            plt.yticks(ticks=[])
    else:
        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_images2(data=origin_data, main_title="origin_data")
print()
show_images2(data=origin_data, main_title="a0s01_data", a=0, s=0.1)
show_images2(data=origin_data, main_title="a0s1_data", a=0, s=1)
show_images2(data=origin_data, main_title="a0s10_data", a=0, s=10)
show_images2(data=origin_data, main_title="a0s40_data", a=0, s=40)
print()
show_images2(data=origin_data, main_title="a10s01_data", a=10, s=0.1)
show_images2(data=origin_data, main_title="a10s1_data", a=10, s=1)
show_images2(data=origin_data, main_title="a10s10_data", a=10, s=10)
show_images2(data=origin_data, main_title="a10s40_data", a=10, s=40)
print()
show_images2(data=origin_data, main_title="a100s01_data", a=100, s=0.1)
show_images2(data=origin_data, main_title="a100s1_data", a=100, s=1)
show_images2(data=origin_data, main_title="a100s10_data", a=100, s=10)
show_images2(data=origin_data, main_title="a100s40_data", a=100, s=40)
print()
show_images2(data=origin_data, main_title="a1000s01_data", a=1000, s=0.1)
show_images2(data=origin_data, main_title="a1000s1_data", a=1000, s=1)
show_images2(data=origin_data, main_title="a1000s10_data", a=1000, s=10)
show_images2(data=origin_data, main_title="a1000s40_data", a=1000, s=40)
print()
show_images2(data=origin_data, main_title="a10000s01_data", a=10000, s=0.1)
show_images2(data=origin_data, main_title="a10000s1_data", a=10000, s=1)
show_images2(data=origin_data, main_title="a10000s10_data", a=10000, s=10)
show_images2(data=origin_data, main_title="a10000s40_data", a=10000, s=40)

Built for developers, by developers.

Whether you're building a simple prototype or a business-critical product, Heroku's fully-managed platform gives you the simplest path to delivering apps quickly — using the tools and languages you already love!

Learn More

DEV Community

ElasticTransform in PyTorch (4)

Built for developers, by developers.

Top comments (0)

Okay