FragFake is a dataset for edited-image detection using Vision-Language Models (VLMs). It contains four groups of examples—Gemini-IG, GoT, MagicBrush, and UltraEdit—each with two difficulty levels: easy and hard. The dataset was created by Vincent-HKUSTGZ and was last updated on July 31, 2025.
Use Cases
- Benchmarking VLM performance on edited-image detection based on the described difficulty levels.
- Training models to distinguish between easy and hard edited images as described.
- Evaluating detection algorithms across different editing techniques (Gemini-IG, GoT, MagicBrush, UltraEdit).
- Studying the robustness of VLMs against image manipulations of varying complexity.
Strengths
- The dataset is structured with four distinct example groups, providing variety.
- It includes two difficulty levels (easy and hard) for controlled evaluation.
- A sampling policy is described to address privacy and content leakage concerns.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- Vincent-HKUSTGZ
- Freshness
- Last updated 2025-07-31 20:56:09; freshness should be verified.