RandAugment in PyTorch (1)

#python #pytorch #randaugment #v2

Buy Me a Coffee☕

*Memos:

My post explains RandAugment() about magnitude argument.
My post explains OxfordIIITPet().

RandAugment() can randomly augment an image as the alternative of AutoAugment() as shown below:

*Memos:

The 1st argument for initialization is num_ops(Optional-Default:2-Type:int). *It must be 0 <= x.
The 2nd argument for initialization is magnitude(Optional-Default:9-Type:int or tuple/list(int or float)): *Memos:
- It must be 0 <= x and 0 < num_magnitude_bins.
The 3rd argument for initialization is num_magnitude_bins(Optional-Default:31-Type:int). *It must be 1 <= x.
The 4th argument for initialization is interpolation(Optional-Default:InterpolationMode.NEAREST-Type:InterpolationMode). *If the input is a tensor, only InterpolationMode.NEAREST, InterpolationMode.BILINEAR can be set to it.
The 5th 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 augmenting an image.
- A tuple/list must be the 1D with 1 or 3 elements.
- If all values are x <= 0, it's black.
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 RandAugment
from torchvision.transforms.functional import InterpolationMode

ra = RandAugment()
ra = RandAugment(num_ops=2, magnitude=9, num_magnitude_bins=31,
                 interpolation = InterpolationMode.NEAREST,
                 fill=None)
ra
# RandAugment(interpolation=InterpolationMode.NEAREST,
#             num_ops=2, magnitude=9, num_magnitude_bins=31)

ra.num_ops
# 2

ra.magnitude
# 9

ra.num_magnitude_bins
# 31

ra.interpolation
# <InterpolationMode.NEAREST: 'nearest'>

print(ra.fill)
# None

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

no0_data = OxfordIIITPet( # `no` is num_ops.
    root="data",
    transform=RandAugment(num_ops=0)
)

no1_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1)
)

no2_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=2)
)

no5_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=5)
)

no10_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=10)
)

no25_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=25)
)

no50_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=50)
)

no100_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=100)
)

no500_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=500)
)

no1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000)
)

no0m30_data = OxfordIIITPet( # `m` is magnitude.
    root="data",
    transform=RandAugment(num_ops=0, magnitude=30)
)

no1m30_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1, magnitude=30)
)

no2m30_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=2, magnitude=30)
)

no5m30_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=5, magnitude=30)
)

no10m30_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=10, magnitude=30)
)

no25m30_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=25, magnitude=30)
)

no0nmb1000_data = OxfordIIITPet( # `nmb` is num_magnitude_bins.
    root="data",
    transform=RandAugment(num_ops=0, num_magnitude_bins=1000)
)

no1nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1, num_magnitude_bins=1000)
)

no2nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=2, num_magnitude_bins=1000)
)

no5nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=5, num_magnitude_bins=1000)
)

no10nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=10, num_magnitude_bins=1000)
)

no25nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=25, num_magnitude_bins=1000)
)

no50nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=50, num_magnitude_bins=1000)
)

no100nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=100, num_magnitude_bins=1000)
)

no500nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=500, num_magnitude_bins=1000)
)

no1000nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1000, num_magnitude_bins=1000)
)

no0m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=0, magnitude=999, num_magnitude_bins=1000)
)

no1m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=1, magnitude=999, num_magnitude_bins=1000)
)

no2m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=2, magnitude=999, num_magnitude_bins=1000)
)

no5m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=5, magnitude=999, num_magnitude_bins=1000)
)

no10m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=10, magnitude=999, num_magnitude_bins=1000)
)

no25m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=25, magnitude=999, num_magnitude_bins=1000)
)

no50m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=50, magnitude=999, num_magnitude_bins=1000)
)

no100m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=100, magnitude=999, num_magnitude_bins=1000)
)

no500m999nmb1000_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=500, magnitude=999, num_magnitude_bins=1000)
)

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

no25f150_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=25, fill=150)
)

no25f160_32_240_data = OxfordIIITPet(
    root="data",
    transform=RandAugment(num_ops=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=no0_data, main_title="no0_data")
show_images1(data=no1_data, main_title="no1_data")
show_images1(data=no2_data, main_title="no2_data")
show_images1(data=no5_data, main_title="no5_data")
show_images1(data=no10_data, main_title="no10_data")
show_images1(data=no25_data, main_title="no25_data")
show_images1(data=no50_data, main_title="no50_data")
show_images1(data=no100_data, main_title="no100_data")
show_images1(data=no500_data, main_title="no500_data")
show_images1(data=no1000_data, main_title="no1000_data")
print()
show_images1(data=no0m30_data, main_title="no0m30_data")
show_images1(data=no1m30_data, main_title="no1m30_data")
show_images1(data=no2m30_data, main_title="no2m30_data")
show_images1(data=no5m30_data, main_title="no5m30_data")
show_images1(data=no10m30_data, main_title="no10m30_data")
show_images1(data=no25m30_data, main_title="no25m30_data")
print()
show_images1(data=no0nmb1000_data, main_title="no0nmb1000_data")
show_images1(data=no1nmb1000_data, main_title="no1nmb1000_data")
show_images1(data=no2nmb1000_data, main_title="no2nmb1000_data")
show_images1(data=no5nmb1000_data, main_title="no5nmb1000_data")
show_images1(data=no10nmb1000_data, main_title="no10nmb1000_data")
show_images1(data=no25nmb1000_data, main_title="no25nmb1000_data")
show_images1(data=no50nmb1000_data, main_title="no50nmb1000_data")
show_images1(data=no100nmb1000_data, main_title="no100nmb1000_data")
show_images1(data=no500nmb1000_data, main_title="no500nmb1000_data")
show_images1(data=no1000nmb1000_data, main_title="no1000nmb1000_data")
print()
show_images1(data=no0m999nmb1000_data, main_title="no0m999nmb1000_data")
show_images1(data=no1m999nmb1000_data, main_title="no1m999nmb1000_data")
show_images1(data=no2m999nmb1000_data, main_title="no2m999nmb1000_data")
show_images1(data=no5m999nmb1000_data, main_title="no5m999nmb1000_data")
show_images1(data=no10m999nmb1000_data, main_title="no10m999nmb1000_data")
show_images1(data=no25m999nmb1000_data, main_title="no25m999nmb1000_data")
show_images1(data=no50m999nmb1000_data, main_title="no50m999nmb1000_data")
show_images1(data=no100m999nmb1000_data, main_title="no100m999nmb1000_data")
show_images1(data=no500m999nmb1000_data, main_title="no500m999nmb1000_data")
show_images1(data=no1000m999nmb1000_data, main_title="no1000m999nmb1000_data")
print()
show_images1(data=no25f150_data, main_title="no25f150_data")
show_images1(data=no25f160_32_240_data, main_title="no25f160_32_240_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="no0_data", no=0)
show_images2(data=origin_data, main_title="no1_data", no=1)
show_images2(data=origin_data, main_title="no2_data", no=2)
show_images2(data=origin_data, main_title="no5_data", no=5)
show_images2(data=origin_data, main_title="no10_data", no=10)
show_images2(data=origin_data, main_title="no25_data", no=25)
show_images2(data=origin_data, main_title="no50_data", no=50)
show_images2(data=origin_data, main_title="no100_data", no=100)
show_images2(data=origin_data, main_title="no500_data", no=500)
show_images2(data=origin_data, main_title="no1000_data", no=1000)
print()
show_images2(data=origin_data, main_title="no0m30_data", no=0, m=30)
show_images2(data=origin_data, main_title="no1m30_data", no=1, m=30)
show_images2(data=origin_data, main_title="no2m30_data", no=2, m=30)
show_images2(data=origin_data, main_title="no5m30_data", no=5, m=30)
show_images2(data=origin_data, main_title="no10m30_data", no=10, m=30)
show_images2(data=origin_data, main_title="no25m30_data", no=25, m=30)
print()
show_images2(data=origin_data, main_title="no0nmb1000_data", no=0,
             nmb=1000)
show_images2(data=origin_data, main_title="no1nmb1000_data", no=1, 
             nmb=1000)
show_images2(data=origin_data, main_title="no2nmb1000_data", no=2, 
             nmb=1000)
show_images2(data=origin_data, main_title="no5nmb1000_data", no=5, 
             nmb=1000)
show_images2(data=origin_data, main_title="no10nmb1000_data", no=10, 
             nmb=1000)
show_images2(data=origin_data, main_title="no25nmb1000_data", no=25, 
             nmb=1000)
show_images2(data=origin_data, main_title="no50nmb1000_data", no=50, 
             nmb=1000)
show_images2(data=origin_data, main_title="no100nmb1000_data", no=100, 
             nmb=1000)
show_images2(data=origin_data, main_title="no500nmb1000_data", no=500, 
             nmb=1000)
show_images2(data=origin_data, main_title="no1000nmb1000_data", no=1000, 
             nmb=1000)
print()
show_images2(data=origin_data, main_title="no0m999nmb1000_data", no=0, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no1m999nmb1000_data", no=1, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no2m999nmb1000_data", no=2, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no5m999nmb1000_data", no=5, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no10m999nmb1000_data", no=10, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no25m999nmb1000_data", no=25, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no50m999nmb1000_data", no=50, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no100m999nmb1000_data", no=100, 
             m=999, nmb=1000)
show_images2(data=origin_data, main_title="no500m999nmb1000_data", no=500,
             m=999, 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="no25f150_data", no=25, f=150)
show_images2(data=origin_data, main_title="no25f160_32_240_data", no=25,
             f=[160, 32, 240])