data/README.md

# Data

This folder documents the dataset layout for the Smart Plate computer vision workflow.

The original dataset should not be committed to GitHub because image datasets are usually
large and may have separate license or usage terms.

## Source

- Dataset: Kaggle Indian Food Images Dataset
- Classes: 80 Indian food categories
- Images: 4,000 total
- Images per class: 50

## Expected Raw Data Layout

Place downloaded images under `data/raw/` or update the notebook/config path to match your
local dataset location.

```text
data/raw/
`-- Indian Food Images/
    `-- Indian Food Images/
        |-- biryani/
        |   |-- image_1.jpg
        |   `-- image_2.jpg
        |-- jalebi/
        `-- ...
```

## Processed Data

Use `data/processed/` for generated CSV files, encoded labels, train/test split files, or
cached preprocessing outputs.

The current notebook builds the dataset from image folders and performs preprocessing in
memory using HuggingFace Datasets and torchvision transforms.