RandomCrop in PyTorch (1)

Buy Me a Coffee☕

*Memos:

• My post explains RandomCrop() about padding, fill and padding_mode argument.
• My post explains RandomCrop() about pad_if_needed argument.
• My post explains RandomResizedCrop() about size argument (1).
• My post explains CenterCrop().
• My post explains FiveCrop().
• My post explains Pad().
• My post explains OxfordIIITPet().

RandomCrop() can randomly crop an image as shown below. *It's about size argument:

*Memos:

• The 1st argument for initialization is size(Required-Type:int or tuple/list(int) or size()): *Memos:
  • It's [height, width].
  • It must be 1 <= x.
  • A tuple/list must be the 1D with 1 or 2 elements.
  • A single value(int or tuple/list(int)) means [size, size].
• The 2nd argument for initialization is padding(Optional-Default:None-Type:int or tuple/list(int)): *Memos:
  • It's [left, top, right, bottom] which can be converted from [left-right, top-bottom] or [left-top-right-bottom].
  • A tuple/list must be the 1D with 1, 2 or 4 elements.
  • A single value(int or tuple/list(int)) means [padding, padding, padding, padding].
  • Double values(tuple/list(int)) means [padding[0], padding[1], padding[0], padding[1]].
• The 3rd argument for initialization is pad_if_needed(Optional-Default:False-Type:bool):
  • If it's False and size is smaller than an original image or the padded image by padding, there is error.
  • If it's True and size is smaller than an original image or the padded image by padding, there is no error, then the image is randomly padded to become size.
• 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 an image is positively padded.
  • A tuple/list must be the 1D with 1 or 3 elements.
  • If all values are x <= 0, it's black.
  • If all values are 255 <= x, it's white.
• The 5th argument for initialization is padding_mode(Optional-Default:'constant'-Type:str). *'constant', 'edge', 'reflect' or 'symmetric' can be set to it.
• The 1st argument is img(Required-Type:PIL Image or tensor(int/float/complex/bool)): *Memos:
  • A tensor must be 2D or more D.
  • 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 RandomCrop

rc = RandomCrop(size=100)
rc = RandomCrop(size=100, padding=None, pad_if_needed=False, fill=0,
                padding_mode='constant')
rc
# RandomCrop(size=(100, 100), pad_if_needed=False, fill=0, 
#            padding_mode=constant)

rc.size
# (100, 100)

print(rc.padding)
# None

rc.pad_if_needed
# False

rc.fill
# 0

rc.padding_mode
# 'constant'

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

s500_394origin_data = OxfordIIITPet( # `s` is size.
    root="data",
    transform=RandomCrop(size=[500, 394])
)

s300_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=300)
    # transform=RandomCrop(size=[300, 300])
)

s200_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=200)
)

s100_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=100)
)

s50_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=50)
)

s10_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=10)
)

s1_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=1)
)

s200_300_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=[200, 300])
)

s300_200_data = OxfordIIITPet(
    root="data",
    transform=RandomCrop(size=[300, 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 in range(1, 6):
        plt.subplot(1, 5, i)
        plt.imshow(X=data[0][0])
    plt.tight_layout()
    plt.show()

plt.figure(figsize=[7, 9])
plt.title(label="s500_394origin_data", fontsize=14)
plt.imshow(X=origin_data[0][0])
show_images1(data=s500_394origin_data, main_title="s500_394origin_data")
show_images1(data=s300_data, main_title="s300_data")
show_images1(data=s200_data, main_title="s200_data")
show_images1(data=s100_data, main_title="s100_data")
show_images1(data=s50_data, main_title="s50_data")
show_images1(data=s10_data, main_title="s10_data")
show_images1(data=s1_data, main_title="s1_data")
show_images1(data=s200_300_data, main_title="s200_300_data")
show_images1(data=s300_200_data, main_title="s300_200_data")

# ↓ ↓ ↓ ↓ ↓ ↓ The code below is identical to the code above. ↓ ↓ ↓ ↓ ↓ ↓
def show_images2(data, main_title=None, s=None, p=None,
                 pin=False, f=0, pm='constant'):
    plt.figure(figsize=[10, 5])
    plt.suptitle(t=main_title, y=0.8, fontsize=14)
    im = data[0][0]
    if not s:
        s = [im.size[1], im.size[0]]
    for i in range(1, 6):
        plt.subplot(1, 5, i)
        rc = RandomCrop(size=s, padding=p, pad_if_needed=pin,
                        fill=f, padding_mode=pm)
        plt.imshow(X=rc(im))
    plt.tight_layout()
    plt.show()

plt.figure(figsize=[7, 9])
plt.title(label="s500_394origin_data", fontsize=14)
plt.imshow(X=origin_data[0][0])
show_images2(data=origin_data, main_title="s500_394origin_data")
# ↑ show_images2(data=origin_data, main_title="s500_394origin_data",
#                s=[500, 394])
show_images2(data=origin_data, main_title="s300_data", s=300)
show_images2(data=origin_data, main_title="s200_data", s=200)
show_images2(data=origin_data, main_title="s100_data", s=100)
show_images2(data=origin_data, main_title="s50_data", s=50)
show_images2(data=origin_data, main_title="s10_data", s=10)
show_images2(data=origin_data, main_title="s1_data", s=1)
show_images2(data=origin_data, main_title="s200_300_data", s=[200, 300])
show_images2(data=origin_data, main_title="s300_200_data", s=[300, 200])