patpragman / data_manufacturer

Image Dataset Resizer

This Python script is designed to process a dataset stored in a CSV file, containing image paths and manually entered scores. The script resizes the images and copies them to a different folder based on specified conditions.

Usage

1. Clone the repository and navigate to the script's directory.
2. Run the script, providing necessary arguments:
   • --random: (optional) Randomly sample data.
   • --balance: (optional) Balance the dataset.
   • --frac: (optional) Fraction of data to keep in random sampling (default is 0.25).
   • --path: (optional) Path to the CSV file containing image paths and scores (default is "../../adventures_in_elodea/data_scores.csv").
   • --seed: (optional) Seed for random operations (default is 42).

Example:

python data_manufacturer.py --random --balance --frac 0.2 --path path/to/data.csv --seed 123

--random: (optional) Randomly sample data.
--balance: (optional) Balance the dataset.
--frac: (optional) Fraction of data to keep in random sampling (default is 0.25).
--path: (optional) Path to the CSV file containing image paths and scores (default is "../../adventures_in_elodea/data_scores.csv").
--seed: (optional) Seed for random operations (default is 42).

Output

Resized images are stored in folders based on specified criteria:

"data_{size}" folders contain resized images.
Images are organized into "entangled" and "not_entangled" subfolders based on scores.

Notes

Ambiguous cases with scores 1 or 2 are excluded.
Image augmentation techniques are applied during resizing.