| import torch | |
| import torchvision | |
| from torch import nn | |
| def create_effnetb4_model(num_classes: int = 101, seed: int = 42): | |
| """ | |
| Creates an EfficientNet-B4 feature extractor model and its preprocessing transforms. | |
| Args: | |
| num_classes (int, optional): Number of output classes. Defaults to 101 (Food-101 dataset). | |
| seed (int, optional): Random seed for reproducibility. Defaults to 42. | |
| Returns: | |
| model (torch.nn.Module): EfficientNet-B4 feature extractor model. | |
| transforms (torchvision.transforms.Compose): Corresponding image transforms. | |
| """ | |
| # Use pretrained EfficientNet-B4 weights | |
| weights = torchvision.models.EfficientNet_B4_Weights.DEFAULT | |
| transforms = weights.transforms() | |
| # Initialize model | |
| model = torchvision.models.efficientnet_b4(weights=weights) | |
| # Freeze feature extractor layers | |
| for param in model.parameters(): | |
| param.requires_grad = False | |
| # Set seed for reproducibility | |
| torch.manual_seed(seed) | |
| # Replace classifier head | |
| model.classifier = nn.Sequential( | |
| nn.Dropout(p=0.3, inplace=True), | |
| nn.Linear(in_features=1792, out_features=num_classes) | |
| ) | |
| return model, transforms |