Content:
The dataset includes 400 scanned document images of automatically generated invoices, printed, stamped, and scanned at 200 dpi resolution. Documents contain color logos and texts, and stamps of various shapes and colors, some overlapping with signatures or text. Some documents have multiple stamps or none. Ground truth is provided as binary images with stamp stroke masks for pixel-wise evaluation.
Folders include:
- Scans: Stamped genuine document scans.
- Ground-truth-maps: Maps defining stamp regions.
- Ground-truth-pixel: Pixel-level ground truth.
- Info: Text files with information for each file, including:
- Signature [0|1]: Signature presence.
- TextOverlap [0|1]: Stamp overlap with printed text.
- NumStamps [0|…|n]: Number of stamps.
- BwStamp [1|…|n]: Stamp color (black or colored).