PentimentoA steganalysis corpus with its licences attached

10,000 cover photographs and 344,357 matched stego pairs, every image carrying its own licence and every file its own checksum.

Get it

What is in it

Limitations

You are allowed to redistribute it

Built from permissively licensed Wikimedia Commons photographs. The standard research corpora in this field either carry no readable licence or forbid publishing anything derived from them.

The credit line travels with the pixels

5,453 of the covers require attribution, and their stego derivatives inherit it. Every sample carries its cover's licence, artist and source URL, so one arm is enough to comply.

Pairs that differ only in the payload

Both halves of every pair come off the same encoder. Where a tool rewrites the file, its clean half is written by that same tool.

Verifiable byte for byte

Every generator is seeded, so the same covers give the same arms. Every file has a sha256 and every sample records how much actually changed, so the corpus can be checked rather than taken on trust.

Start here

If you want to	Read
Know what this is	What it is
Download it	Get it
Know what is inside	What is in it
Train on it	Loading and splitting
Publish a number from it	Limitations
Comply with the licence	Licence and attribution