RandMark: On Random Watermarking of Visual Foundation Models

Researchers propose RandMark, a new method for watermarking visual foundation models to protect intellectual property rights. The approach uses a small encoder-decoder network to embed random digital watermarks into internal representations, enabling ownership verification with low false detection rates.

• →RandMark enables ownership verification of valuable visual foundation models through random watermark embedding.
• →The method uses a small encoder-decoder network to watermark internal representations rather than direct model parameters.
• →Experimental results show low probability of both false detection and false misdetection.
• →The approach addresses intellectual property protection for computationally expensive AI models.
• →Random watermarking makes ownership statistics detectable in functional copies of watermarked models.