JoyCN/ai-generated-ecommerce-images

---
license: cc0-1.0
task_categories:
- image-classification
language:
- zh
- en
tags:
- ai-generated
- e-commerce
- fraud-detection
- synthetic
- imagefolder
size_categories:
- 1K<n<10K
---

# AI-Generated E-Commerce Fraud Detection Images

## Overview

This dataset contains **4,424 AI-generated images** simulating common after-sales fraud
scenarios in cross-border e-commerce. It is intended for training and evaluating
AI-generated image detection models.

## Dataset Structure

| Category | Description |
|---|---|
| `damaged_electronics` | Electronics with fake damage |
| `damaged_phone_screen` | Phone screens with fabricated cracks |
| `damaged_product` | General products with simulated damage |
| `fake_logistics_screenshot` | Fabricated logistics/shipping screenshots |
| `fragile_broken` | Broken fragile items (glass, ceramics, etc.) |
| `measurement_mismatch` | Items with size/spec discrepancies |
| `missing_parts` | Products with missing components |
| `package_damage` | Damaged packaging |
| `quality_issue` | Quality defects and surface flaws |
| `skin_allergy` | Skin reactions (redness, rash) |
| `supplement_spill` | Leaked/spilled supplement products |
| `wrong_item` | Incorrect items received |

## Data Source

All images were generated using **Google Gemini** via the text-to-image API.
No real user photos or personally identifiable information are included.

## Intended Use

- Training binary classifiers to detect AI-generated images in after-sales claims
- Evaluating multimodal VLM fine-tuning pipelines (e.g., Gemma-3, InternVL)
- Research on synthetic image detection in e-commerce contexts

## License

**CC0 1.0 Universal (Creative Commons Zero)**

To the extent possible under law, the creator waives all copyright and related
rights to this dataset via CC0. Where copyright does not exist (e.g., purely
AI-generated works), this dataset is dedicated to the Public Domain.

See [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/) for details.

## Creator

[Martin (JoyCN)](https://huggingface.co/JoyCN)