Briefly describe what topics you learned when you created this code sample: from…
- User
Briefly describe what topics you learned when you created this code sample:
from google.colab import drive drive.mount('/content/drive')
Load the dataset from the zip file
import zipfile with zipfile.ZipFile('/content/drive/MyDrive/archive (1).zip', 'r') as zip_ref: zip_ref.extractall('/content/dataset')
import torch import torch.nn as nn import torch.optim as optim from torch.utils.data import DataLoader from torchvision.transforms import Compose, ToTensor, Normalize, RandomRotation, RandomHorizontalFlip from torchvision.datasets import ImageFolder
from matplotlib import pyplot as plt import numpy as np import os import random
Parameters
IMG_HEIGHT = 48 IMG_WIDTH = 48 batch_size = 32 epochs = 50 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
Directories
import torchvision train_data_dir = '/content/dataset/train' validation_data_dir = '/content/dataset/test'
mean = [0.5, 0.5, 0.5] std = [0.5, 0.5, 0.5]
Define the data transforms for train and test sets
data_transforms = { 'train': torchvision.transforms.Compose([ torchvision.transforms.Resize(256), # Resize the image to 256x256 pixels torchvision.transforms.RandomCrop(224), # Crop a random 224x224 patch from the image torchvision.transforms.RandomRotation(30), # Rotate the image randomly by up to 30 degrees torchvision.transforms.RandomHorizontalFlip(), # Flip the image horizontally with a probability of 0.5 torchvision.transforms.ToTensor(), # Convert the image to a PyTorch tensor torchvision.transforms.Normalize(mean, std) # Normalize the image using the mean and standard deviation ]), 'test': torchvision.transforms.Compose([ torchvision.transforms.Resize(256), # Resize the image to 256x256 pixels torchvision.transforms.CenterCrop(224), # Crop the center 224x224 patch from the image torchvision.transforms.ToTensor(), # Convert the image to a PyTorch tensor torchvision.transforms.Normalize(mean, std) # Normalize the image using the mean and standard deviation ]) }
Datasets
train_dataset = ImageFolder(train_data_dir, data_transforms['train']) test_dataset = ImageFolder(validation_data_dir, data_transforms['test'])
DataLoaders
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True) test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)
class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise']
import matplotlib.pyplot as plt import numpy as np
batch = next(iter(train_loader)) images, labels = batch
images = images.numpy() / 2 + 0.5
fig, axes = plt.subplots(nrows=3, ncols=3, figsize=(8, 8))
for i, ax in enumerate(axes.flat): ax.imshow(np.transpose(images[i], (1, 2, 0))) ax.set_title(f"Label: {class_labels[labels[i]]}")
plt.show()
#Custom model
import torch.nn.functional as F
Model architecture
class EmotionModel(nn.Module): def init(self): super(EmotionModel, self).init() self.conv1 = nn.Conv2d(3, 32, kernel_size=3) self.conv2 = nn.Conv2d(32, 64, kernel_size=3) self.pool2 = nn.MaxPool2d(2) self.drop2 = nn.Dropout(0.1) self.conv3 = nn.Conv2d(64, 128, kernel_size=3) self.pool3 = nn.MaxPool2d(2) self.drop3 = nn.Dropout(0.1) self.conv4 = nn.Conv2d(128, 256, kernel_size=3) self.pool4 = nn.MaxPool2d(2) self.drop4 = nn.Dropout(0.1) self.fc1 = nn.Linear(4096, 512) self.drop5 = nn.Dropout(0.2) self.fc2 = nn.Linear(512, 7)
def forward(self, x): x = F.relu(self.conv1(x)) x = self.drop2(self.pool2(F.relu(self.conv2(x)))) x = self.drop3(self.pool3(F.relu(self.conv3(x)))) x = self.drop4(self.pool4(F.relu(self.conv4(x)))) # print(x.size()) # Add this line to print the size of the tensor x = x.view(-1, 4096) x = F.relu(self.fc1(x)) x = self.drop5(x) x = self.fc2(x) return xmodel = EmotionModel().to(device) criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters())
Train the model
for epoch in range(epochs): model.train() running_loss = 0.0 for i, data in enumerate(train_loader): inputs, labels = data inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f"Epoch {epoch+1}/{epochs}, Loss: {running_loss / len(train_loader):.4f}")Save the model
torch.save(model.state_dict(), 'emotion_detection_model_50epochs.pth')
Test the model
model.eval() correct = 0 total = 0 with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item()
print(f"Accuracy: {100 * correct / total:.2f}%")
Plot confusion matrix
from sklearn.metrics import confusion_matrix import seaborn as sns
y_true = [] y_pred = [] with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) y_true.extend(labels.cpu().numpy()) y_pred.extend(predicted.cpu().numpy())
cm = confusion_matrix(y_true, y_pred) class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise'] sns.heatmap(cm, annot=True, xticklabels=class_labels, yticklabels=class_labels) plt.show()
Display a few test images with original and predicted labels
n_display = 6 fig, axes = plt.subplots(1, n_display, figsize=(15, 3)) for i in range(n_display): index = random.randint(0, len(test_dataset)) image, label = test_dataset[index] image = image.to(device) output = model(image.unsqueeze(0)) _, prediction = torch.max(output.data, 1) orig_label = class_labels[label] pred_label = class_labels[prediction.item()]
img = image.cpu().numpy().transpose((1, 2, 0)) img = img * 0.5 + 0.5 # Un-normalize img = np.clip(img, 0, 1) axes[i].imshow(img, cmap="gray") axes[i].set_title(f"Original: {orig_label}\nPredicted: {pred_label}") axes[i].axis("off")plt.show()
#ResNet
import torch import torch.nn as nn import torch.optim as optim from torchvision.models import resnet18
Define the pre-trained ResNet18 model
model = resnet18(pretrained=True) num_ftrs = model.fc.in_features model.fc = nn.Linear(num_ftrs, 7)
Move the model to the GPU if available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model = model.to(device)
Define the loss function and optimizer
criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters())
Train the model
for epoch in range(epochs): model.train() running_loss = 0.0 for i, data in enumerate(train_loader): inputs, labels = data inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f"Epoch {epoch+1}/{epochs}, Loss: {running_loss / len(train_loader):.4f}")Save the model
torch.save(model.state_dict(), 'emotion_detection_model_resnet.pth')
model = resnet18(pretrained=True) num_ftrs = model.fc.in_features model.fc = nn.Linear(num_ftrs, 7)
Load the pre-trained weights from a file
weights_path = "emotion_detection_model_resnet.pth" model.load_state_dict(torch.load(weights_path)) model=model.to(device)
Test the model
model.eval() correct = 0 total = 0 with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item()
print(f"Accuracy: {100 * correct / total:.2f}%")
Plot confusion matrix
from sklearn.metrics import confusion_matrix import seaborn as sns
y_true = [] y_pred = [] with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) y_true.extend(labels.cpu().numpy()) y_pred.extend(predicted.cpu().numpy())
cm = confusion_matrix(y_true, y_pred) class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise'] sns.heatmap(cm, annot=True, xticklabels=class_labels, yticklabels=class_labels) plt.show()
Display a few test images with original and predicted labels
n_display = 6 fig, axes = plt.subplots(1, n_display, figsize=(15, 3)) for i in range(n_display): index = random.randint(0, len(test_dataset)) image, label = test_dataset[index] image = image.to(device) output = model(image.unsqueeze(0)) _, prediction = torch.max(output.data, 1) orig_label = class_labels[label] pred_label = class_labels[prediction.item()]
img = image.cpu().numpy().transpose((1, 2, 0)) img = img * 0.5 + 0.5 # Un-normalize img = np.clip(img, 0, 1) axes[i].imshow(img, cmap="gray") axes[i].set_title(f"Original: {orig_label}\nPredicted: {pred_label}") axes[i].axis("off")plt.show()
#Mobilenet
from torchvision.models import mobilenet_v2
Define the pre-trained MobileNetV2 model
model = mobilenet_v2(pretrained=True) num_ftrs = model.classifier[1].in_features model.classifier[1] = nn.Linear(num_ftrs, 7)
Move the model to the GPU if available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model = model.to(device)
Define the loss function and optimizer
criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters())
Train the model
for epoch in range(epochs): model.train() running_loss = 0.0 for i, data in enumerate(train_loader): inputs, labels = data inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f"Epoch {epoch+1}/{epochs}, Loss: {running_loss / len(train_loader):.4f}")Save the model
torch.save(model.state_dict(), 'emotion_detection_model_wideresnet.pth')
model = mobilenet_v2(pretrained=True) num_ftrs = model.classifier[1].in_features model.classifier[1] = nn.Linear(num_ftrs, 7) model=model.to(device)
Load the pre-trained weights from a file
weights_path = "/content/emotion_detection_model_wideresnet.pth" model.load_state_dict(torch.load(weights_path)) model=model.to(device)
Test the model
model.eval() correct = 0 total = 0 with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item()
print(f"Accuracy: {100 * correct / total:.2f}%")
Plot confusion matrix
from sklearn.metrics import confusion_matrix import seaborn as sns
y_true = [] y_pred = [] with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) y_true.extend(labels.cpu().numpy()) y_pred.extend(predicted.cpu().numpy())
cm = confusion_matrix(y_true, y_pred) class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise'] sns.heatmap(cm, annot=True, xticklabels=class_labels, yticklabels=class_labels) plt.show()
#Squeezenet
import torch import torch.nn as nn import torch.optim as optim from torchvision.models import squeezenet1_0
Define the pre-trained SqueezeNet model
model = squeezenet1_0(pretrained=True) num_ftrs = model.classifier[1].in_channels model.classifier[1] = nn.Conv2d(num_ftrs, 7, kernel_size=(1,1), stride=(1,1))
Move the model to the GPU if available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model = model.to(device)
Define the loss function and optimizer
criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters())
Train the model
for epoch in range(epochs): model.train() running_loss = 0.0 for i, data in enumerate(train_loader): inputs, labels = data inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f"Epoch {epoch+1}/{epochs}, Loss: {running_loss / len(train_loader):.4f}")Save the model
torch.save(model.state_dict(), 'emotion_detection_model_squeezenet.pth')
model = squeezenet1_0(pretrained=True) num_ftrs = model.classifier[1].in_channels model.classifier[1] = nn.Conv2d(num_ftrs, 7, kernel_size=(1,1), stride=(1,1)) model=model.to(device)
Load the pre-trained weights from a file
weights_path = "/content/emotion_detection_model_squeezenet.pth" model.load_state_dict(torch.load(weights_path)) model=model.to(device)
Test the model
model.eval() correct = 0 total = 0 with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item()
print(f"Accuracy: {100 * correct / total:.2f}%")
Plot confusion matrix
from sklearn.metrics import confusion_matrix import seaborn as sns
y_true = [] y_pred = [] with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) y_true.extend(labels.cpu().numpy()) y_pred.extend(predicted.cpu().numpy())
cm = confusion_matrix(y_true, y_pred) class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise'] sns.heatmap(cm, annot=True, xticklabels=class_labels, yticklabels=class_labels) plt.show()
#Shufflenet
Define the pre-trained ShuffleNetV2 model
from torchvision.models import shufflenet_v2_x1_0 model = shufflenet_v2_x1_0(pretrained=True) num_ftrs = model.fc.in_features model.fc = nn.Linear(num_ftrs, 7)
Move the model to the GPU if available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model = model.to(device)
Define the loss function and optimizer
criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters())
Train the model
num_epochs=50 for epoch in range(num_epochs): model.train() running_loss = 0.0 for i, data in enumerate(train_loader): inputs, labels = data inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() running_loss += loss.item() print(f"Epoch {epoch+1}/{num_epochs}, Loss: {running_loss / len(train_loader):.4f}")Save the model
torch.save(model.state_dict(), 'emotion_detection_model_shufflenet.pth')
Test the model
model.eval() correct = 0 total = 0 with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item()
print(f"Accuracy: {100 * correct / total:.2f}%")
Plot confusion matrix
from sklearn.metrics import confusion_matrix import seaborn as sns
y_true = [] y_pred = [] with torch.no_grad(): for data in test_loader: images, labels = data images, labels = images.to(device), labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) y_true.extend(labels.cpu().numpy()) y_pred.extend(predicted.cpu().numpy())
cm = confusion_matrix(y_true, y_pred) class_labels = ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise'] sns.heatmap(cm, annot=True, xticklabels=class_labels, yticklabels=class_labels) plt.show()
#Tensorboard
import torch import torch.nn as nn import torch.optim as optim from torch.profiler import profile, record_function, ProfilerActivity from torchvision.models import mobilenet_v2 import torchvision import torchvision.transforms as transforms from torch.utils.data import DataLoader from torch.utils.tensorboard import SummaryWriter
from google.colab import drive drive.mount('/content/drive')
Load the dataset from the zip file
import zipfile with zipfile.ZipFile('/content/drive/MyDrive/archive.zip', 'r') as zip_ref: zip_ref.extractall('/content/dataset')
import torch import torch.nn as nn import torch.optim as optim from torch.utils.data import DataLoader from torchvision.transforms import Compose, ToTensor, Normalize, RandomRotation, RandomHorizontalFlip from torchvision.datasets import ImageFolder
from matplotlib import pyplot as plt import numpy as np import os import random
Parameters
IMG_HEIGHT = 48 IMG_WIDTH = 48 batch_size = 32 epochs = 50 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
Directories
import torchvision train_data_dir = '/content/dataset/train' validation_data_dir = '/content/dataset/test'
mean = [0.5, 0.5, 0.5] std = [0.5, 0.5, 0.5]
Define the data transforms for train and test sets
data_transforms = { 'train': torchvision.transforms.Compose([ torchvision.transforms.Resize(256), # Resize the image to 256x256 pixels torchvision.transforms.RandomCrop(224), # Crop a random 224x224 patch from the image torchvision.transforms.RandomRotation(30), # Rotate the image randomly by up to 30 degrees torchvision.transforms.RandomHorizontalFlip(), # Flip the image horizontally with a probability of 0.5 torchvision.transforms.ToTensor(), # Convert the image to a PyTorch tensor torchvision.transforms.Normalize(mean, std) # Normalize the image using the mean and standard deviation ]), 'test': torchvision.transforms.Compose([ torchvision.transforms.Resize(256), # Resize the image to 256x256 pixels torchvision.transforms.CenterCrop(224), # Crop the center 224x224 patch from the image torchvision.transforms.ToTensor(), # Convert the image to a PyTorch tensor torchvision.transforms.Normalize(mean, std) # Normalize the image using the mean and standard deviation ]) }
Datasets
train_dataset = ImageFolder(train_data_dir, data_transforms['train']) test_dataset = ImageFolder(validation_data_dir, data_transforms['test'])
DataLoaders
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True) test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)
!pip install tensorboard !pip install torch-tb-profiler
def profile_model(model, weights_path, log_dir): model.load_state_dict(torch.load(weights_path)) model = model.to(device)
writer = SummaryWriter(log_dir=log_dir) with torch.profiler.profile( schedule=torch.profiler.schedule(wait=0, warmup=2, active=6, repeat=1), on_trace_ready=torch.profiler.tensorboard_trace_handler(log_dir), # Pass log_dir instead of writer record_shapes=True, profile_memory=True, with_stack=True, with_flops=True, activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA], ) as prof: for i, (inputs, labels) in enumerate(test_loader): inputs, labels = inputs.to(device), labels.to(device) with record_function("forward"): outputs = model(inputs) loss = criterion(outputs, labels) if i >= 10: break prof.step() writer.close()from torchvision.models import mobilenet_v2, shufflenet_v2_x1_0, resnet18, squeezenet1_0
resnet = resnet18(pretrained=True) num_ftrs = resnet.fc.in_features resnet.fc = nn.Linear(num_ftrs, 7)
mobilenet = mobilenet_v2(pretrained=True) num_ftrs = mobilenet.classifier[1].in_features mobilenet.classifier[1] = nn.Linear(num_ftrs, 7)
squeezenet = squeezenet1_0(pretrained=True) num_ftrs = squeezenet.classifier[1].in_channels squeezenet.classifier[1] = nn.Conv2d(num_ftrs, 7, kernel_size=(1,1), stride=(1,1))
shufflenet = shufflenet_v2_x1_0(pretrained=True) num_ftrs = shufflenet.fc.in_features shufflenet.fc = nn.Linear(num_ftrs, 7)
criterion = nn.CrossEntropyLoss()
models_and_paths = [ (mobilenet, "/content/drive/MyDrive/emotion_detection_model_mobilenet.pth", "runs/profiler_mobilenet"), (shufflenet, "/content/drive/MyDrive/emotion_detection_model_shufflenet.pth", "runs/profiler_shufflenet"), (resnet, "/content/drive/MyDrive/emotion_detection_model_resnet.pth", "runs/profiler_resnet"), (squeezenet, "/content/drive/MyDrive/emotion_detection_model_squeezenet.pth", "runs/profiler_squeezenet"), ]
for model, weights_path, log_dir in models_and_paths: profile_model(model, weights_path, log_dir)
%load_ext tensorboard %tensorboard --logdir runs
- Assistant
This code sample demonstrates how to:
- Load a dataset from a zip file
- Preprocess and augment image data
- Create custom models, such as Convolutional Neural Networks (CNNs) for emotion detection using PyTorch
- Train and evaluate the custom models
- Implement pretrained models (ResNet, MobileNetV2, SqueezeNet, and ShuffleNetV2) for emotion detection using transfer learning
- Save and load trained model weights
- Calculate model accuracy and display confusion matrix
- Visualize original and predicted labels for test images
- Profile models using TensorBoard
- Run and visualize Tensorboard for profiling and performance analysis.