This project aims to create a Python package for generating diverse and enriched image datasets from a small original dataset using three augmentation families:

1. Traditional Augmentation: Flips, rotations, scaling, cropping, color jitter, etc., implemented via Albumentations.
2. Neural Style Transfer (NST): Applies artistic/domain-specific textures from style images, implemented with PyTorch + pre-trained fast NST models.
3. Patch Mixing: Combines regions from different images (CutMix, MixUp) to boost structural diversity.

• Produce lightweight, diverse datasets for small-data training scenarios.
• Allow custom combinations of techniques per batch.

• Gradio-based UI: For interactive usage, allowing users to upload base datasets and optional style images, choose augmentation pipelines and parameters, and preview generated samples in real-time.
• Python API & CLI: For batch automation.
• Export: To standard dataset formats (COCO, ImageFolder, etc.).
• Diversity Scoring: (LPIPS, FID) with visual reports.

1. User uploads original images.
2. Selects techniques (checklist) and parameters (sliders for rotation, blend ratio, style strength).
3. Previews augmented images instantly.
4. Clicks "Generate & Download" to export the batch.