Super Kai (Kazuya Ito)
Posted on Mar 16
RandAugment in PyTorch (3)

#python #pytorch #randaugment #v2
*Memos:
My post explains RandAugment() about no arguments and fill argument.
My post explains RandAugment() about num_ops and fill argument.
My post explains RandAugment() about num_magnitude_bins argument.
RandAugment() can randomly augment an image as shown below. *It's about magnitude and fill argument:
from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import RandAugment
from torchvision.transforms.functional import InterpolationMode

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

m0_data = OxfordIIITPet( # `m` is magnitude.
    root="data",
    transform=RandAugment(magnitude=0)
)

m1_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=1)
)

m2_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=2)
)

m5_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=5)
)

m10_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=10)
)

m25_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=25)
)

no1000m0_data = OxfordIIITPet( # `no` is num_ops.
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=0)
)

no1000m1_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=1)
)

no1000m2_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=2)
)

no1000m5_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=5)
)

no1000m10_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=10)
)

no1000m25_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=25)
)

m0nmb1000_data = OxfordIIITPet( # `nmb` is num_magnitude_bins.
    root="data",
    transform=RandAugment(magnitude=0, num_magnitude_bins=1000)
)

m1nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=1, num_magnitude_bins=1000)
)

m2nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=2, num_magnitude_bins=1000)
)

m5nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=5, num_magnitude_bins=1000)
)

m10nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=10, num_magnitude_bins=1000)
)

m25nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=25, num_magnitude_bins=1000)
)

m50nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=50, num_magnitude_bins=1000)
)

m100nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=100, num_magnitude_bins=1000)
)

m500nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=500, num_magnitude_bins=1000)
)

m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=999, num_magnitude_bins=1000)
)

no1000m0nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=0, num_magnitude_bins=1000)
)

no1000m1nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=1, num_magnitude_bins=1000)
)

no1000m2nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=2, num_magnitude_bins=1000)
)

no1000m5nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=5, num_magnitude_bins=1000)
)

no1000m10nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=10, num_magnitude_bins=1000)
)

no1000m25nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=25, num_magnitude_bins=1000)
)

no1000m50nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=50, num_magnitude_bins=1000)
)

no1000m100nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=100, num_magnitude_bins=1000)
)

no1000m500nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=500, num_magnitude_bins=1000)
)

no1000m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, magnitude=999, num_magnitude_bins=1000)
)

m25fgray_data = OxfordIIITPet( # `f` is fill.
    root="data",
    transform=RandAugment(magnitude=25, fill=150)
    # transform=RandAugment(magnitude=25, fill=[150])
)

m25fpurple_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(magnitude=25, 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")
print()
show_images1(data=m0_data, main_title="m0_data")
show_images1(data=m1_data, main_title="m1_data")
show_images1(data=m2_data, main_title="m2_data")
show_images1(data=m5_data, main_title="m5_data")
show_images1(data=m10_data, main_title="m10_data")
show_images1(data=m25_data, main_title="m25_data")
print()
show_images1(data=no1000m0_data, main_title="no1000m0_data")
show_images1(data=no1000m1_data, main_title="no1000m1_data")
show_images1(data=no1000m2_data, main_title="no1000m2_data")
show_images1(data=no1000m5_data, main_title="no1000m5_data")
show_images1(data=no1000m10_data, main_title="no1000m10_data")
show_images1(data=no1000m25_data, main_title="no1000m25_data")
print()
show_images1(data=m0nmb1000_data, main_title="m0nmb1000_data")
show_images1(data=m1nmb1000_data, main_title="m1nmb1000_data")
show_images1(data=m2nmb1000_data, main_title="m2nmb1000_data")
show_images1(data=m5nmb1000_data, main_title="m5nmb1000_data")
show_images1(data=m10nmb1000_data, main_title="m10nmb1000_data")
show_images1(data=m25nmb1000_data, main_title="m25nmb1000_data")
show_images1(data=m50nmb1000_data, main_title="m50nmb1000_data")
show_images1(data=m100nmb1000_data, main_title="m100nmb1000_data")
show_images1(data=m500nmb1000_data, main_title="m500nmb1000_data")
show_images1(data=m999nmb1000_data, main_title="m999nmb1000_data")
print()
show_images1(data=no1000m0nmb1000_data, main_title="no1000m0nmb1000_data")
show_images1(data=no1000m1nmb1000_data, main_title="no1000m1nmb1000_data")
show_images1(data=no1000m2nmb1000_data, main_title="no1000m2nmb1000_data")
show_images1(data=no1000m5nmb1000_data, main_title="no1000m5nmb1000_data")
show_images1(data=no1000m10nmb1000_data, main_title="no1000m10nmb1000_data")
show_images1(data=no1000m25nmb1000_data, main_title="no1000m25nmb1000_data")
show_images1(data=no1000m50nmb1000_data, main_title="no1000m50nmb1000_data")
show_images1(data=no1000m100nmb1000_data, main_title="no1000m100nmb1000_data")
show_images1(data=no1000m500nmb1000_data, main_title="no1000m500nmb1000_data")
show_images1(data=no1000m999nmb1000_data, main_title="no1000m999nmb1000_data")
print()
show_images1(data=m25fgray_data, main_title="m25fgray_data")
show_images1(data=m25fpurple_data, main_title="m25fpurple_data")

# ↓ ↓ ↓ ↓ ↓ ↓ The code below is identical to the code above. ↓ ↓ ↓ ↓ ↓ ↓
def show_images2(data, main_title=None, no=2, m=9, nmb=31,
                 ip=InterpolationMode.NEAREST, f=None):
    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)
            ra = RandAugment(num_ops=no, magnitude=m,
                             num_magnitude_bins=nmb,
                             interpolation=ip, fill=f)
            plt.imshow(X=ra(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="m0_data", m=0)
show_images2(data=origin_data, main_title="m1_data", m=1)
show_images2(data=origin_data, main_title="m2_data", m=2)
show_images2(data=origin_data, main_title="m5_data", m=5)
show_images2(data=origin_data, main_title="m10_data", m=10)
show_images2(data=origin_data, main_title="m25_data", m=25)
print()
show_images2(data=origin_data, main_title="no1000m0_data", no=1000, m=0)
show_images2(data=origin_data, main_title="no1000m1_data", no=1000, m=1)
show_images2(data=origin_data, main_title="no1000m2_data", no=1000, m=2)
show_images2(data=origin_data, main_title="no1000m5_data", no=1000, m=5)
show_images2(data=origin_data, main_title="no1000m10_data", no=1000, m=10)
show_images2(data=origin_data, main_title="no1000m25_data", no=1000, m=25)
print()
show_images2(data=origin_data, main_title="m0nmb1000_data", m=0, nmb=1000)
show_images2(data=origin_data, main_title="m1nmb1000_data", m=1, nmb=1000)
show_images2(data=origin_data, main_title="m2nmb1000_data", m=2, nmb=1000)
show_images2(data=origin_data, main_title="m5nmb1000_data", m=5, nmb=1000)
show_images2(data=origin_data, main_title="m10nmb1000_data", m=10, nmb=1000)
show_images2(data=origin_data, main_title="m25nmb1000_data", m=25, nmb=1000)
show_images2(data=origin_data, main_title="m50nmb1000_data", m=50, nmb=1000)
show_images2(data=origin_data, main_title="m100nmb1000_data", m=100, nmb=1000)
show_images2(data=origin_data, main_title="m500nmb1000_data", m=500, nmb=1000)
show_images2(data=origin_data, main_title="m999nmb1000_data", m=999, nmb=1000)
print()
show_images2(data=origin_data, main_title="no1000m0nmb1000_data", no=1000, m=0,
             nmb=1000)
show_images2(data=origin_data, main_title="no1000m1nmb1000_data", no=1000, m=1,
             nmb=1000)
show_images2(data=origin_data, main_title="no1000m2nmb1000_data", no=1000, m=2,
             nmb=1000)
show_images2(data=origin_data, main_title="no1000m5nmb1000_data", no=1000, m=5,
             nmb=1000)
show_images2(data=origin_data, main_title="no1000m10nmb1000_data", no=1000,
             m=10, nmb=1000)
show_images2(data=origin_data, main_title="no1000m25nmb1000_data", no=1000,
             m=25, nmb=1000)
show_images2(data=origin_data, main_title="no1000m50nmb1000_data", no=1000,
             m=50, nmb=1000)
show_images2(data=origin_data, main_title="no1000m100nmb1000_data", no=1000,
             m=100, nmb=1000)
show_images2(data=origin_data, main_title="no1000m500nmb1000_data", no=1000,
             m=500, nmb=1000)
show_images2(data=origin_data, main_title="no1000m999nmb1000_data", no=1000,
             m=999, nmb=1000)
print()
show_images2(data=origin_data, main_title="m25fgray_data", m=25, f=150)
show_images2(data=origin_data, main_title="m25fpurple_data", m=25,
             f=[160, 32, 240])
The Next Generation Developer Platform

Coherence is the first Platform-as-a-Service you can control. Unlike "black-box" platforms that are opinionated about the infra you can deploy, Coherence is powered by CNC, the open-source IaC framework, which offers limitless customization.
Learn more
DEV Community

RandAugment in PyTorch (3)

The Next Generation Developer Platform

Top comments (0)

Join us for AWS Security LIVE!

Read next

How to Check the Bitrate of an MP3 File Using Python

Fixing ModuleNotFoundError: No module named 'moviepy.editor' in Python

How to Extract Audio from an MP4 Video and Save as MP3 Using MoviePy

Deployable On-Premises RAG

Okay
