Food101 Dataset Preprocessing

This project preprocesses the Food101 dataset for image classification. It includes:

• Resize and convert all images to 512×512 RGB .jpg format
• Filter selected classes (e.g., pizza, sushi, burger, etc.)
• Uniform random sampling (300 samples per class)
• Generate train.csv, val.csv, and test.csv with image paths and labels
• Support for torchvision.datasets.ImageFolder and custom PyTorch Dataset using CSV annotations

Content

• CustomDataset_Loader.py: Resize raw Food101 images to 512×512 RGB, filter selected classes with uniform sampling, create train/val/test CSV annotations, optionally zip CSVs, and provide a PyTorch Dataset class for CSV-based loading with transforms and DataLoaders.

• ImageFolder_Loader.py: Split a folder of class-labeled images into train/val/test folders with an 80/10/10 ratio, apply image transforms, create PyTorch ImageFolder datasets and DataLoaders, and print dataset statistics.

• projects_structure.dm: A visualization of each project directory structure.

Important Note

This repository does not include code to download the Food101 dataset. The preprocessing workflow is a virtual exercise designed for practice.

How to Download the Food101 Dataset

Download the official Food101 dataset here:

• Food101 dataset (official)

After downloading and extracting the dataset, place the images directory in the expected folder structure before running preprocessing.

---

requirements.txt

• pandas
• numpy
• pillow
• torch
• torchvision

Setup

Install required dependencies:

pip install pandas numpy pillow torch torchvision