102,028 images are grouped into 11,142 subsets, each containing an original image and manipulated derivatives. The dataset was created by Silvan Heller of the University of Basel for research on media derivation and tampering detection. It was sourced from a large community of image manipulation enthusiasts.
Use Cases
- Training models to detect image manipulations based on the original-derivative pairs.
- Benchmarking automated tampering detection methods against human perception.
- Studying the types of edits made by a community of manipulation enthusiasts.
Strengths
- Contains 102,028 total images, providing substantial scale.
- Organized into 11,142 subsets with clear original and derivative relationships.
- Sourced from an active community, likely capturing diverse manipulation techniques.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Last update date is unknown; freshness unverified.
- Data may reflect bias inherent to the specific online community it was sourced from.
Provenance
- Source
- University of Basel, sourced from an online community of image manipulation enthusiasts.
- Collection Method
- Gathered from a large community of image manipulation enthusiasts.