EfficientNet-B7 Pizza/Steak/Sushi Classifier

I fine-tuned a pre-trained EfficientNet-B7 model to classify images into three categories: pizza, steak, and sushi.

Model Details

• Architecture: torchvision.models.efficientnet_b7
• Weights: EfficientNet_B7_Weights.DEFAULT
• Modifications: I froze all the base feature layers.

Training Procedure

I trained the model for 10 epochs using the Adam optimizer.

• Batch Size: 32
• Learning Rate: 0.001
• Loss Function: CrossEntropyLoss
• Transforms: I used the automatic transforms provided by the default EfficientNet-B7 weights.
• Hardware: Trained using cuda (if available) with a set manual seed of 37 for reproducibility.

Dataset

I used a 20% subset of a pizza, steak, and sushi dataset. The data was split into train and test directories.

Evaluation Results

Accuracy and Loss Curves

Over the 10 epochs, the training and testing loss steadily decreased, with the testing loss ending near a highly impressive 0.15. The testing accuracy consistently outperformed the training accuracy throughout, finishing near a perfect 99%.

Confusion Matrix

This model achieved the best performance across all three classes on the test set, making almost no errors:

• Pizza: 46 correct, 0 misclassified.
• Steak: 57 correct, 0 misclassified as pizza, 1 misclassified as sushi.
• Sushi: 46 correct, 0 misclassified.

Most Confident Wrong Predictions

I plotted the single instance where the model was incorrect. The model only made one mistake in the entire test set, predicting a steak dish as sushi with a relatively low confidence of 0.56.

How to use

import torch
import torchvision

# I loaded the model architecture
weights = torchvision.models.EfficientNet_B7_Weights.DEFAULT
model = torchvision.models.efficientnet_b7(weights=weights)

# I modified the classifier
model.classifier = torch.nn.Sequential(
    torch.nn.Dropout(p=0.2, inplace=True),
    torch.nn.Linear(in_features=2560, out_features=3, bias=True),
)

# I loaded the saved weights
model.load_state_dict(torch.load("EfficientNet_B7_20percent.pth", map_location="cpu"))
model.eval()