FoodVision Mini is a fast and efficient food classification model built using EfficientNetB2. It is trained on a subset of the Food-101 dataset, focusing on three classes: Pizza, Steak, and Sushi.
- Model: EfficientNetB2 (pretrained on ImageNet with a modified classifier head)
- Classes: 3 (Pizza, Steak, Sushi)
- Accuracy: ~96.88% on the test set
- Model Size: ~30 MB – optimized for mobile deployment
- Fast Inference: Designed for rapid predictions with minimal latency.
- High Accuracy: Achieves high performance on a curated 3-class dataset.
- Deployment Ready: Packaged with a Gradio demo for interactive testing and easy sharing.
- Data Handling: Custom data loaders built with PyTorch’s
ImageFolderfor training and testing. - Training: Implements training and evaluation loops with detailed tracking of loss and accuracy.
- Inference: A dedicated prediction function processes images, measures inference time, and outputs class probabilities.
- Deployment: Includes a Gradio interface demo and is packaged for deployment as a standalone app.