Winner of the JPEG Trust Watermarking Grand Challenge
Trufo was named one of the three winning teams in the JPEG Trust Grand Challenge for image watermarking.
We are excited to announce that Trufo has been named one of the three winning teams in the JPEG Trust Grand Challenge, evaluated based on a new JPEG Trust Watermarking Benchmark, under the submission name byh. Presentations will be held at the IEEE International Conference on Image Processing (ICIP) in Tampere, Finland later this year. A representative from Trufo will be in attendance.
This challenge is aligned with JPEG Trust Part 3: Media Asset Watermarking (ISO/IEC 21617). The watermark must embed a 100-bit payload, and is scored on both visual imperceptibility and robustness across a number of attacks (geometric edits, signal processing, compression codecs, generative AI modifications).
For the challenge, Trufo submitted a modified version of its 2024 watermark, which relies entirely on traditional CPU-based techniques. The watermark was adjusted to accommodate 100 bits instead of the original 52 bits.
Since 2024, neural network watermarking techniques have improved considerably. Our 2024 watermark, that once was far stronger than any other offering on the market, is now just neck-and-neck with 2026 SotA neural network implementations. Of course, we are not unfamiliar with AI/ML, and are planning to release a new line of neural network watermarks ahead of the EU AI Act deadline on August 2nd.