How do I use a saved model in Pytorch to predict the label of a never before seen image?

Question

I have been trying to use my pretrained model to predict the label on a never before seen image. I have trained a CNN to classify flowers of 5 types using the Kaggle flower recognition dataset. I so far have trained my model to 97% accuracy and saved the model to a directory. I now want to download any image of a flower from those types and be able to use this pretrained model to predict the label. So far this is my code: (A code review for all of this would be very helpful as this is my first ever project)

This is my CNN model that I train:

from multiprocessing import freeze_support

import torch
from torch import nn
from torch.autograd import Variable
from torch.utils.data import DataLoader, Sampler
from torchvision import datasets
from torchvision.transforms import transforms
from torch.optim import Adam

import matplotlib.pyplot as plt
import numpy as np

# Hyperparameters.
num_epochs = 20
num_classes = 5
batch_size = 100
learning_rate = 0.001
num_of_workers = 5

DATA_PATH_TRAIN = 'C:\\Users\Aeryes\PycharmProjects\simplecnn\images\\train\\'
DATA_PATH_TEST = 'C:\\Users\Aeryes\PycharmProjects\simplecnn\images\\test\\'
MODEL_STORE_PATH = 'C:\\Users\Aeryes\PycharmProjects\simplecnn\model'

trans = transforms.Compose([
    transforms.RandomHorizontalFlip(),
    transforms.Resize(32),
    transforms.CenterCrop(32),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5),(0.5, 0.5, 0.5))
    ])

# Flowers dataset.
train_dataset = datasets.ImageFolder(root=DATA_PATH_TRAIN, transform=trans)
test_dataset = datasets.ImageFolder(root=DATA_PATH_TEST, transform=trans)

# Create custom random sampler class to iter over dataloader.
train_loader = DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True, num_workers=num_of_workers)
test_loader = DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False, num_workers=num_of_workers)

# CNN we are going to implement.
class Unit(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(Unit, self).__init__()

        self.conv = nn.Conv2d(in_channels=in_channels, kernel_size=3, out_channels=out_channels, stride=1, padding=1)
        self.bn = nn.BatchNorm2d(num_features=out_channels)
        self.relu = nn.ReLU()

    def forward(self, input):
        output = self.conv(input)
        output = self.bn(output)
        output = self.relu(output)

        return output


class CNNet(nn.Module):
    def __init__(self, num_class):
        super(CNNet, self).__init__()

        # Create 14 layers of the unit with max pooling in between
        self.unit1 = Unit(in_channels=3, out_channels=32)
        self.unit2 = Unit(in_channels=32, out_channels=32)
        self.unit3 = Unit(in_channels=32, out_channels=32)

        self.pool1 = nn.MaxPool2d(kernel_size=2)

        self.unit4 = Unit(in_channels=32, out_channels=64)
        self.unit5 = Unit(in_channels=64, out_channels=64)
        self.unit6 = Unit(in_channels=64, out_channels=64)
        self.unit7 = Unit(in_channels=64, out_channels=64)

        self.pool2 = nn.MaxPool2d(kernel_size=2)

        self.unit8 = Unit(in_channels=64, out_channels=128)
        self.unit9 = Unit(in_channels=128, out_channels=128)
        self.unit10 = Unit(in_channels=128, out_channels=128)
        self.unit11 = Unit(in_channels=128, out_channels=128)

        self.pool3 = nn.MaxPool2d(kernel_size=2)

        self.unit12 = Unit(in_channels=128, out_channels=128)
        self.unit13 = Unit(in_channels=128, out_channels=128)
        self.unit14 = Unit(in_channels=128, out_channels=128)

        self.avgpool = nn.AvgPool2d(kernel_size=4)

        # Add all the units into the Sequential layer in exact order
        self.net = nn.Sequential(self.unit1, self.unit2, self.unit3, self.pool1, self.unit4, self.unit5, self.unit6
                                 , self.unit7, self.pool2, self.unit8, self.unit9, self.unit10, self.unit11, self.pool3,
                                 self.unit12, self.unit13, self.unit14, self.avgpool)

        self.fc = nn.Linear(in_features=128, out_features=num_class)

    def forward(self, input):
        output = self.net(input)
        output = output.view(-1, 128)
        output = self.fc(output)
        return output


# Check if gpu support is available
cuda_avail = torch.cuda.is_available()

# Create model, optimizer and loss function
model = CNNet(num_classes)

# if cuda is available, move the model to the GPU
if cuda_avail:
    model.cuda()

# Define the optimizer and loss function
optimizer = Adam(model.parameters(), lr=0.0001, weight_decay=0.0001)
loss_fn = nn.CrossEntropyLoss()

def save_models(epoch):
    torch.save(model.state_dict(), f"flowermodel_{epoch}.model")
    print("Checkpoint saved")

def test():
    model.eval()
    test_acc = 0.0
    for i, (images, labels) in enumerate(test_loader):

        if cuda_avail:
            images = Variable(images.cuda())
            labels = Variable(labels.cuda())

        # Predict classes using images from the test set
        outputs = model(images)
        _, prediction = torch.max(outputs.data, 1)

        test_acc += torch.sum(prediction == labels.data).float()

    # Compute the average acc and loss over all 10000 test images
    test_acc = test_acc / 4242 * 100

    return test_acc


def train(num_epoch):
    best_acc = 0.0

    for epoch in range(num_epoch):
        model.train()
        train_acc = 0.0
        train_loss = 0.0
        for i, (images, labels) in enumerate(train_loader):
            # Move images and labels to gpu if available
            if cuda_avail:
                images = Variable(images.cuda())
                labels = Variable(labels.cuda())

            # Clear all accumulated gradients
            optimizer.zero_grad()
            # Predict classes using images from the test set
            outputs = model(images)
            # Compute the loss based on the predictions and actual labels
            loss = loss_fn(outputs, labels)
            # Backpropagate the loss
            loss.backward()

            # Adjust parameters according to the computed gradients
            optimizer.step()

            train_loss += loss.cpu().data[0] * images.size(0)
            _, prediction = torch.max(outputs.data, 1)

            train_acc += torch.sum(prediction == labels.data).float()

        # Call the learning rate adjustment function
        #adjust_learning_rate(epoch)

        # Compute the average acc and loss over all 50000 training images
        train_acc = train_acc / 4242 * 100
        train_loss = train_loss / 8484

        # Evaluate on the test set
        test_acc = test()

        # Save the model if the test acc is greater than our current best
        if test_acc > best_acc:
            save_models(epoch)
            best_acc = test_acc

        # Print the metrics
        print(f"Epoch {epoch + 1}, Train Accuracy: {train_acc} , TrainLoss: {train_loss} , Test Accuracy: {test_acc}")

if __name__ == '__main__':
    freeze_support()
    train(num_epochs)

This is my image loader to view preprocessed images:

from multiprocessing import freeze_support

import torch
from torch import nn
import torchvision
from torch.autograd import Variable
from torch.utils.data import DataLoader, Sampler
from torchvision import datasets
from torchvision.transforms import transforms
from torch.optim import Adam

import matplotlib.pyplot as plt
import numpy as np
import PIL

num_classes = 5
batch_size = 100
num_of_workers = 5

DATA_PATH_TRAIN = 'C:\\Users\Aeryes\PycharmProjects\simplecnn\images\\train'
DATA_PATH_TEST = 'C:\\Users\Aeryes\PycharmProjects\simplecnn\images\\test'

trans = transforms.Compose([
    transforms.RandomHorizontalFlip(),
    transforms.Resize(32),
    transforms.CenterCrop(32),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5),(0.5, 0.5, 0.5))
    ])

train_dataset = datasets.ImageFolder(root=DATA_PATH_TRAIN, transform=trans)
train_loader = DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True, num_workers=num_of_workers)

def imshow(img):
    img = img / 2 + 0.5 # unnormalize
    #npimg = img.numpy()
    plt.imshow(np.transpose(img[0].numpy(), (1, 2, 0)))
    plt.show()

def main():
    # get some random training images
    dataiter = iter(train_loader)
    images, labels = dataiter.next()

    # show images
    imshow(images)

if __name__ == "__main__":
    main()

So far this is what I have for my classify new image file which will classify new images:

from multiprocessing import freeze_support

import torch
from torch import nn
from torch.autograd import Variable
from torch.utils.data import DataLoader, Sampler
from torchvision import datasets
from torchvision.transforms import transforms
from torch.optim import Adam

import matplotlib.pyplot as plt
import numpy as np

def classify_new_image():
    # Classify a new image using a pretrained model from the above training.

    # Location of the image we will classify.
    IMG_PATH = "C:\\Users\\Aeryes\\PycharmProjects\\simplecnn\\images\\pretrain_classify\\"



    # Pre-processing the new image using transform.
    min_img_size = 32
    trans = transforms.Compose([transforms.Resize(min_img_size),
                                         transforms.CenterCrop(32),
                                         transforms.ToTensor(),
                                         transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                                              std=[0.229, 0.224, 0.225])])


# Picture dataset.
classify_dataset = datasets.ImageFolder(root=IMG_PATH, transform=trans)
# Create custom random sampler class to iter over dataloader.
classify_loader = DataLoader(dataset=classify_dataset, batch_size=1, shuffle=True, num_workers=5)

# Check if gpu support is available
cuda_avail = torch.cuda.is_available()

model = torch.load('C:\\Users\\Aeryes\\PycharmProjects\\simplecnn\\src\\flowermodel_20.tar')['state_dict']

# if cuda is available, move the model to the GPU
if cuda_avail:
    model.cuda()

if __name__ == "__main__":
    classify_new_image()

A big issue that I am also facing is making sense of the outputs. I printed the prediction variable from the CNN models and it gave me a tensor of numbers ranging from 0-4, which I presume is the 5 classes I have in my data folders. If anyone can help me make sense of this a I would be very grateful.

Here is my direct question: How do I use my pre-trained model to predict never before seen images of flowers?

Answer 1

You can do like this for a single image,

import torch
from torchvision.transforms import transforms
from PIL import Image
from cnn_main import CNNet
from pathlib import Path

model = CNNet(5)
checkpoint = torch.load(Path('C:/Users/Aeryes/PycharmProjects/simplecnn/src/19.model'))
model.load_state_dict(checkpoint)
trans = transforms.Compose([
    transforms.RandomHorizontalFlip(),
    transforms.Resize(32),
    transforms.CenterCrop(32),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5),(0.5, 0.5, 0.5))
    ])

image = Image.open(Path('C:/Users/Aeryes/PycharmProjects/simplecnn/images/pretrain_classify/rose_classify.jpg'))

input = trans(image)

input = input.view(1, 3, 32,32)

output = model(input)

prediction = int(torch.max(output.data, 1)[1].numpy())
print(prediction)

if (prediction == 0):
    print ('daisy')
if (prediction == 1):
    print ('dandelion')
if (prediction == 2):
    print ('rose')
if (prediction == 3):
    print ('sunflower')
if (prediction == 4):
    print ('tulip')