MobileNet is a family of deep learning models designed specifically for efficient execution on mobile and embedded devices. The primary goal of MobileNet architectures is to achieve a balance between model size, computational complexity, and accuracy, making them well-suited for applications where computational resources and power consumption are limited.
The primary innovation in MobileNet is the use of depthwise separable convolutions, which reduce the computational cost and model size significantly.
- A standard convolution operation is divided into two parts:
- Depthwise Convolution: Applies a single convolutional filter per input channel.
- Pointwise Convolution: Uses a 1x1 convolution to combine the outputs of the depthwise convolution.
- MobileNetV1: Introduced the concept of depthwise separable convolutions, resulting in significant reductions in the number of parameters and computational cost compared to traditional convolutional networks.
- MobileNetV2: Improved upon V1 by introducing inverted residuals and linear bottlenecks, enhancing both efficiency and performance.
- MobileNetV3: Further optimized using a combination of platform-aware neural architecture search (NAS) and a novel network structure, balancing accuracy and efficiency even more effectively.
MobileNet models include a width multiplier parameter (α), which scales the number of channels in each layer, allowing for the adjustment of the model size and computational cost.
- α values range from 0 to 1, where lower values reduce the model size and computational load.
MobileNet allows the input image resolution to be reduced, further decreasing the computational complexity.
MobileNets are widely used in mobile and embedded vision applications, such as image classification, object detection, and face recognition, due to their efficient architecture.
MobileNet models are highly efficient convolutional neural networks optimized for mobile and embedded devices. They achieve significant reductions in model size and computational complexity through the use of depthwise separable convolutions, making them ideal for resource-constrained environments. Various versions of MobileNet (V1, V2, V3) offer different levels of optimization, allowing users to select the appropriate balance of efficiency and performance for their specific application.
This code is a script for training an image classification model MobileNet using PyTorch. The code is structured to facilitate easy training, evaluation, and monitoring of a deep learning model for image classification. It allows for periodic saving of the model's state and provides insights into the model's performance through loss curves and accuracy metrics.
Here is a step-by-step summary of the code:
- Libraries: Import necessary libraries including PyTorch, torchvision, einops, and other utility libraries such as matplotlib and pandas.
import torch
import torch.nn as nn
import torch.nn.functional as F
import matplotlib.pyplot as plt
from einops.layers.torch import Rearrange
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
from torchvision.datasets import ImageFolder
import os
import pandas as pd
from models.models.create_models import ImageClassifier- Set environment variables to avoid certain errors.
os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'- Define the root directory for the dataset and the model name. Specify the number of classes and instantiate the model.
dataset_root = r"C:\dataset"
model_name = "mobilenetv2_100"
no_classes = 11
model = ImageClassifier(no_classes, model_name)- Define training parameters such as image size, number of epochs, and batch size.
image_size = 224
num_epochs = 100
batch_size = 4- Configure the optimizer and the loss function for training.
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
criterion = nn.CrossEntropyLoss()- Define transformations to be applied to the images such as resizing, tensor conversion, and normalization.
transform = transforms.Compose([
transforms.Resize((image_size, image_size)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])- Create datasets and dataloaders for training, validation, and testing.
train_dataset = ImageFolder(root=dataset_root + "/train", transform=transform)
validation_dataset = ImageFolder(root=dataset_root + "/test", transform=transform)
train_dataloader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
validation_dataloader = DataLoader(validation_dataset, batch_size=batch_size, shuffle=True)
test_dataloader = DataLoader(validation_dataset, batch_size=4, shuffle=False)- Train the model for a specified number of epochs, compute training and validation loss, and save the model every 10 epochs.
training_loss_arr = []
validation_loss_arr = []
for epoch in range(num_epochs):
model.train()
total_loss = 0.0
for images, labels in train_dataloader:
optimizer.zero_grad()
outputs = model(images)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
total_loss += loss.item()
avg_train_loss = total_loss / len(train_dataloader)
training_loss_arr.append(avg_train_loss)
model.eval()
total_val_loss = 0.0
for images, labels in validation_dataloader:
with torch.no_grad():
outputs = model(images)
val_loss = criterion(outputs, labels)
total_val_loss += val_loss.item()
avg_val_loss = total_val_loss / len(validation_dataloader)
validation_loss_arr.append(avg_val_loss)
print(f"Epoch {epoch + 1}: Train Loss: {avg_train_loss:.4f}, Val Loss: {avg_val_loss:.4f}")
if (epoch + 1) % 10 == 0:
torch.save(model.state_dict(), "{}_{}.pth".format(model_name, epoch))
print("Model saved successfully.")- Validate the model after every 10 epochs and calculate the accuracy.
total_correct = 0
total_samples = 0
with torch.no_grad():
for images, labels in test_dataloader:
outputs = model(images)
_, predicted = torch.max(outputs.data, 1)
total_samples += labels.size(0)
total_correct += (predicted == labels).sum().item()
accuracy = 100 * total_correct / total_samples
print("Validation Accuracy: {:.2f}%".format(accuracy))- Save the training and validation loss to a CSV file.
df = pd.DataFrame({'train_loss': training_loss_arr, 'val_loss': validation_loss_arr})
df.to_csv('C:/trainandvallossagain.csv', index=False)- Plot the training and validation loss curves and save the plot as an EPS file.
plt.figure(figsize=(8, 4), dpi=300)
plt.plot(training_loss_arr, label="Training Loss")
plt.plot(validation_loss_arr, label="Validation Loss")
plt.xlabel('Epoch', fontsize=12, fontweight='bold')
plt.ylabel('Loss Value', fontsize=12, fontweight='bold')
plt.title('Loss curve', fontsize=14, fontweight='bold')
plt.legend()
plt.savefig('Loss Curve.eps', format='eps', bbox_inches='tight')
plt.show()- Save the final trained model's state dictionary.
torch.save(model.state_dict(), "vit_model.pth")
print("Model saved successfully.")
Explore More! 🚀