MovingMNIST dalam PyTorch

Beli Saya Kopi☕

*Siaran saya menerangkan Moving MNIST.

MovingMNIST() boleh menggunakan Moving MNIST dataset seperti yang ditunjukkan di bawah:

*Memo:

• Argumen pertama ialah root(Required-Type:str or pathlib.Path). *Laluan mutlak atau relatif boleh dilakukan.
• Argumen ke-2 dipecahkan(Optional-Default:None-Type:str): *Memo:
  • Tiada, "kereta api" atau "ujian" boleh ditetapkan padanya.
  • Jika Tiada, kesemua 20 bingkai(imej) setiap video dikembalikan, mengabaikan nisbah_pecah.
• Argumen ke-3 ialah split_ratio(Optional-Default:10-Type:int): *Memo:
  • Jika split ialah "kereta api", data[:, :split_ratio] dikembalikan.
  • Jika belah ialah "ujian", data[:, split_ratio:] dikembalikan.
  • Jika perpecahan Tiada, ia diabaikan. mengabaikan nisbah_pecah.
• Argumen ke-4 ialah transform(Optional-Default:None-Type:callable).
• Argumen ke-5 ialah muat turun(Optional-Default:False-Type:bool): *Memo:
  • Jika Benar, set data dimuat turun dari internet ke akar.
  • Jika ia Benar dan set data sudah dimuat turun, ia akan diekstrak.
  • Jika ia Benar dan set data sudah dimuat turun, tiada apa yang berlaku.
  • Ia sepatutnya Palsu jika set data sudah dimuat turun kerana ia lebih pantas.
  • Anda boleh memuat turun dan mengekstrak set data secara manual dari sini ke mis. data/MovingMNIST/.

from torchvision.datasets import MovingMNIST

all_data = MovingMNIST(
    root="data"
)

all_data = MovingMNIST(
    root="data",
    split=None,
    split_ratio=10,
    download=False,
    transform=None
)

train_data = MovingMNIST(
    root="data",
    split="train"
)

test_data = MovingMNIST(
    root="data",
    split="test"
)

len(all_data), len(train_data), len(test_data)
# (10000, 10000, 10000)

len(all_data[0]), len(train_data[0]), len(test_data[0])
# (20, 10, 10)

all_data
# Dataset MovingMNIST
#     Number of datapoints: 10000
#     Root location: data

all_data.root
# 'data'

print(all_data.split)
# None

all_data.split_ratio
# 10

all_data.download
# <bound method movingmnist.download of dataset movingmnist number datapoints: root location: data>

print(all_data.transform)
# None

from torchvision.datasets import MovingMNIST

import matplotlib.pyplot as plt

plt.figure(figsize=(10, 3))

plt.subplot(1, 3, 1)
plt.title("all_data")
plt.imshow(all_data[0].squeeze()[0])

plt.subplot(1, 3, 2)
plt.title("train_data")
plt.imshow(train_data[0].squeeze()[0])

plt.subplot(1, 3, 3)
plt.title("test_data")
plt.imshow(test_data[0].squeeze()[0])

plt.show()
</bound>

from torchvision.datasets import MovingMNIST

all_data = MovingMNIST(
    root="data",
    split=None
)

train_data = MovingMNIST(
    root="data",
    split="train"
)

test_data = MovingMNIST(
    root="data",
    split="test"
)

def show_images(data, main_title=None):
    plt.figure(figsize=(10, 8))
    plt.suptitle(t=main_title, y=1.0, fontsize=14)
    for i, image in enumerate(data, start=1):
        plt.subplot(4, 5, i)
        plt.tight_layout(pad=1.0)
        plt.title(i)
        plt.imshow(image)
    plt.show()

show_images(data=all_data[0].squeeze(), main_title="all_data")
show_images(data=train_data[0].squeeze(), main_title="train_data")
show_images(data=test_data[0].squeeze(), main_title="test_data")

from torchvision.datasets import MovingMNIST

all_data = MovingMNIST(
    root="data",
    split=None
)

train_data = MovingMNIST(
    root="data",
    split="train"
)

test_data = MovingMNIST(
    root="data",
    split="test"
)

import matplotlib.pyplot as plt

def show_images(data, main_title=None):
    plt.figure(figsize=(10, 8))
    plt.suptitle(t=main_title, y=1.0, fontsize=14)
    col = 5
    for i, image in enumerate(data, start=1):
        plt.subplot(4, 5, i)
        plt.tight_layout(pad=1.0)
        plt.title(i)
        plt.imshow(image.squeeze()[0])
        if i == col:
            break
    plt.show()

show_images(data=all_data, main_title="all_data")
show_images(data=train_data, main_title="train_data")
show_images(data=test_data, main_title="test_data")

from torchvision.datasets import MovingMNIST
import matplotlib.animation as animation

all_data = MovingMNIST(
    root="data"
)

import matplotlib.pyplot as plt
from IPython.display import HTML

figure, axis = plt.subplots()

# ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ `ArtistAnimation()` ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
images = []
for image in all_data[0].squeeze():
    images.append([axis.imshow(image)])
ani = animation.ArtistAnimation(fig=figure, artists=images,
                                interval=100)
# ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ `ArtistAnimation()` ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑

# ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ `FuncAnimation()` ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
# def animate(i):
#     axis.imshow(all_data[0].squeeze()[i])
#
# ani = animation.FuncAnimation(fig=figure, func=animate,
#                               frames=20, interval=100)
# ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ `FuncAnimation()` ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑

# ani.save('result.gif') # Save the animation as a `.gif` file

plt.ioff() # Hide a useless image

# ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ Show animation ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
HTML(ani.to_jshtml()) # Animation operator
# HTML(ani.to_html5_video()) # Animation video
# ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ Show animation ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑

# ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ Show animation ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
# plt.rcParams["animation.html"] = "jshtml" # Animation operator
# plt.rcParams["animation.html"] = "html5" # Animation video
# ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ Show animation ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑

from torchvision.datasets import MovingMNIST
from ipywidgets import interact, IntSlider

all_data = MovingMNIST(
    root="data"
)

import matplotlib.pyplot as plt
from IPython.display import HTML

def func(i):
    plt.imshow(all_data[0].squeeze()[i])

interact(func, i=(0, 19, 1))
# interact(func, i=IntSlider(min=0, max=19, step=1, value=0))
# ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ Set the start value ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑ ↑
plt.show()

Atas ialah kandungan terperinci MovingMNIST dalam PyTorch. Untuk maklumat lanjut, sila ikut artikel berkaitan lain di laman web China PHP!