python - 从 pytorch 访问文件夹中的图像并将其提供给 CNN 预测的推理方法

Question

import torch, torchvision
from torch import nn
from torch import optim
from torchvision.transforms import ToTensor
import torch.nn.functional as F
import matplotlib.pyplot as plt

import requests
from PIL import Image
from io import BytesIO

import copy
from sklearn.metrics import confusion_matrix
import pandas as pd
import numpy as np

import torch.utils.data as data_utils

# indices = torch.arange(1000)
# tr_1k = data_utils.Subset(tr, indices)

from google.colab import drive
drive.mount('/content/gdrive')


numb_batch = 64

T = torchvision.transforms.Compose([
    torchvision.transforms.ToTensor()
])

print('........get MNIST Data............')
train_data = torchvision.datasets.MNIST('mnist_data', train=True, download=True, transform=T)
val_data = torchvision.datasets.MNIST('mnist_data', train=False, download=True, transform=T)

print('slice')
indices_train = torch.arange(1000)
indices_test = torch.arange(1000)
tr_1k = data_utils.Subset(train_data, indices_train)
test_1k = data_utils.Subset(val_data, indices_test)

print('........Assign MNIST data as training and validation dataset...........')
train_dl = torch.utils.data.DataLoader(tr_1k, batch_size = numb_batch)
val_dl = torch.utils.data.DataLoader(test_1k, batch_size = numb_batch)

def create_lenet():
    print('.......create the convolutional neural network..........')
    model = nn.Sequential(
        nn.Conv2d(1, 6, 5, padding=2),
        nn.ReLU(),
        nn.AvgPool2d(2, stride=2),
        nn.Conv2d(6, 16, 5, padding=0),
        nn.ReLU(),
        nn.AvgPool2d(2, stride=2),
        nn.Flatten(),
        nn.Linear(400, 120),
        nn.ReLU(),
        nn.Linear(120, 84),
        nn.ReLU(),
        nn.Linear(84, 10)
    )
    return model

def validate(model, data):
    print('.......validate...........')
    total = 0
    correct = 0
    for i, (images, labels) in enumerate(data):
        # images = images.cuda()
        x = model(images)
        value, pred = torch.max(x,1)
        pred = pred.data.cpu()
        total += x.size(0)
        correct += torch.sum(pred == labels)
    return correct*100./total




def train(numb_epoch=3, lr=1e-3, device="cpu"):
    print('.......train the model................')
    accuracies = []
    cnn = create_lenet().to(device)
    cec = nn.CrossEntropyLoss()
    optimizer = optim.Adam(cnn.parameters(), lr=lr)
    max_accuracy = 0
    for epoch in range(numb_epoch):
        for i, (images, labels) in enumerate(train_dl):
            images = images.to(device)
            labels = labels.to(device)
            optimizer.zero_grad()
            pred = cnn(images)
            loss = cec(pred, labels)
            loss.backward()
            optimizer.step()
        accuracy = float(validate(cnn, val_dl))
        accuracies.append(accuracy)
        if accuracy > max_accuracy:
            best_model = copy.deepcopy(cnn)
            max_accuracy = accuracy
            print("Saving Best Model with Accuracy: ", accuracy)
        print('Epoch:', epoch+1, "Accuracy :", accuracy, '%')
    plt.plot(accuracies)
    return best_model


if torch.cuda.is_available():
    device = torch.device("cuda:0")
else:
    device = torch.device("cpu")
    print("No Cuda Available")

device

print('......start training............')
lenet = train(40, device="cpu")

print('..........save the model...........')
torch.save(lenet.state_dict(), "new.pth")


# lenet = create_lenet().to(device)
# lenet.load_state_dict(torch.load("lenet.pth"))
# lenet.eval()

def predict_dl(model, data):
    print('use the model to predict')
    y_pred = []
    y_true = []
    for i, (images, labels) in enumerate(data):
        # images = images.cuda()
        x = model(images)
        value, pred = torch.max(x, 1)
        pred = pred.data.cpu()
        y_pred.extend(list(pred.numpy()))
        y_true.extend(list(labels.numpy()))
    return np.array(y_pred), np.array(y_true)

y_pred, y_true = predict_dl(lenet, val_dl)

pd.DataFrame(confusion_matrix(y_true, y_pred, labels=np.arange(0,10)))


def inference(path, model, device):
    r = requests.get(path)
    with BytesIO(r.content) as f:
        img = Image.open(f).convert(mode="L")
        img = img.resize((28, 28))
        x = (255 - np.expand_dims(np.array(img), -1))/255.
    with torch.no_grad():
        pred = model(torch.unsqueeze(T(x), axis=0).float().to(device))
        return F.softmax(pred, dim=-1).cpu().numpy()

def inference2(images, model, device):
    with torch.no_grad():
        pred = model(torch.unsqueeze(T(images), axis=0).float().to(device))
        return F.softmax(pred, dim=-1).cpu().numpy()



path = "https://previews.123rf.com/images/aroas/aroas1704/aroas170400068/79321959-handwritten-sketch-black-number-8-on-white-background.jpg"
# path2 = 'https://drive.google.com/file/d/1-t6EEaGFwsq4dsd8UDVVdv4yxCkr7iFW/view?usp=sharing'
path3 = '/content/100images/fake_images-10.png'
r = requests.get(path)
with BytesIO(r.content) as f:
    img = Image.open(f).convert(mode="L")
    img = img.resize((28, 28))
x = (255 - np.expand_dims(np.array(img), -1))/255.

plt.imshow(x.squeeze(-1), cmap="gray")

pred = inference(path, lenet, device=device)
pred_idx = np.argmax(pred)
print(f"Predicted: {pred_idx}, Prob: {pred[0][pred_idx]*100} %")


print('get final fake 100 images')
indices_fake = torch.arange(100)
tr_fake = data_utils.Subset(train_data, indices_fake)
train_dl = torch.utils.data.DataLoader(tr_fake, batch_size = 1)


# '/content/gdrive/MyDrive/2/'
tsr_img1 = torchvision.io.read_image('/content/gdrive/MyDrive/100images/fake_image-10.png')
# for loop to get 100 fake images and cllassify from this model
image_datasets = torchvision.datasets.folder('/content/100images/', transform=T)
dataloadersdataloaders = torch.utils.data.DataLoader(image_datasets, batch_size=1, shuffle=True)

for i, (images,_) in enumerate(tr_fake):
  print(i)
  images = images.view(-1, 784).to(device)
  # with BytesIO(images.content) as f:
  #   img = Image.open(f).convert(mode="L")
  #   img = img.resize((28, 28))
  # x = (255 - np.expand_dims(np.array(images), -1))/255.
  print(images)
  data = images.reshape(-1, 1, 28, 28)
  print(data)
  img = data.resize((28, 28))
  x = (255 - np.expand_dims(np.array(img), -1))/255.
  plt.imshow(x.squeeze(-1) , cmap="gray")

  pred = inference2(images, lenet, device=device)
  pred_idx = np.argmax(pred)
  print(f"Predicted: {pred_idx}, Prob: {pred[0][pred_idx]*100} %")

我有以下代码。在这里，我有一个训练有素的 CNN，它可以预测和分类手写数字。这是使用 mnist 数据集训练的。我想做的是使用其中的推理方法对我自己的数据集进行预测。

我有一个从 mnist 训练集中获取的包含 100 位数字的数据集。我将它们放在 google collab 根位置以及我的 google drive 中。我想获取那些图像（在文件夹中保存为 png s）并通过推理方法运行它们并预测它是什么数字。我无法这样做。我尝试了多种方法未能成功地从文件夹。我不断收到这样的文件夹不存在错误。其次，一旦获取，我需要遍历它们并获得对它们中的每一个的预测。此处的任何帮助将不胜感激。首先获取图像，然后对其进行预测。此处已对从网络上获取的单个图像进行了预测。我需要对其进行修改以预测我们提供的图像。

标签： pythonpytorchconv-neural-networkmnisttorchvision