AI Fraud Detection: Image Pair Datasets

Hello,

I need a dataset of 4,000 image pairs for training an AI model that detects photo manipulation/editing.

Requirements:

Dataset 1: FOOD (2,000 pairs)
- 2,000 original unedited food photos
- 2,000 edited versions of the SAME photos
- Photos should show: restaurant meals, delivery food, groceries, cooked dishes
- Edits should make food look WORSE (for fraud detection): add defects, change colors to look spoiled, add foreign objects, make portions look smaller

Dataset 2: CARS (2,000 pairs)
- 2,000 original unedited car photos
- 2,000 edited versions of the SAME photos
- Photos should show: car exteriors, damage areas, scratches, dents
- Edits should make cars look BETTER (for insurance fraud detection): remove scratches, fix dents, improve paint, hide damage

Technical Specifications:

- Format: JPEG or PNG
- Minimum resolution: 1024x768
- File naming: food_001_original.jpg / food_001_edited.jpg
- Edits must be done using mobile apps (native app, Snapseed, TouchRetouch, Lightroom Mobile, FaceApp, PicsArt, or similar)
- Include a CSV file listing each pair and what edits were applied

Delivery Structure:

/food/original/
/food/edited/
/cars/original/
/cars/edited/
metadata.csv

What I DON'T want:

- AI-generated images (MidJourney, DALL-E, etc.)
- Stock photos with watermarks
- Low quality or blurry images
- Same edit applied to all photos (need variety)

Timeline:

- Flexible, quality over speed.