Binary Image Classification of Pizza and Non-Pizza Dishes

Encompassing approximately 1000 images of pizza and 1000 images of non-pizza dishes, totaling around 2000 images for a binary classification task. All images are rescaled to a maximum side length of 512 pixels. It is a curated subset of the Food-101 dataset, created by nateraw.

Use Cases

• Train a binary image classifier to distinguish pizza from non-pizza dishes using the approximately 2000 labeled images.
• Fine-tune a convolutional neural network on the rescaled 512-pixel images for a simple food recognition task.
• Benchmark lightweight vision models on a balanced dataset of two food categories derived from Food-101.
• Use the pizza and non-pizza image labels to test data augmentation techniques for food image classification.

Strengths

• Contains a balanced set of approximately 1000 images per class, suitable for binary classification.
• All images are preprocessed to a consistent maximum dimension of 512 pixels.
• Derived from the established Food-101 dataset, providing a known provenance.

Limitations

• The dataset size of around 2000 images is relatively small for training deep learning models from scratch.
• As a subset of Food-101, it may inherit any geographic or culinary biases present in the original dataset.
• No information is provided on image resolution, format, or potential label noise.

Provenance

Source

Subset of the Food-101 dataset.

Collection Method

Images were curated and rescaled from the original Food-101 dataset.

Freshness

Last updated on 2022-07-07.