Lightweight-Mark: Rethinking Deep Learning-Based Watermarking

Deep learning-based watermarking models play a crucial role in copyright protection across various applications. However, many high-performance models are limited in practical deployment due to their large number of parameters. Meanwhile, the robustness and invisibility performance of existing lightweight models are unsatisfactory. This presents a pressing need for a watermarking model that combines lightweight capacity with satisfactory performance. Our research identifies a key reason that limits the performance of existing watermarking frameworks: a mismatch between commonly used decoding losses (e.g., mean squared error and binary cross-entropy loss) and the actual decoding goal, leading to parameter redundancy. We propose two innovative solutions: (1) Decoding-oriented surrogate loss (DO), which redesigns the loss function to mitigate the influence of decoding-irrelevant optimization directions; and (2) Detachable projection head (PH), which incorporates a detachable redundant module during training to handle these irrelevant directions and is discarded during inference. Additionally, we propose a novel watermarking framework comprising five submodules, allowing for independent parameter reduction in each component. Our proposed model achieves better efficiency, invisibility, and robustness while utilizing only 2.2\% of the parameters compared to the state-of-the-art frameworks. By improving efficiency while maintaining robust copyright protection, our model is well suited for practical applications in resource-constrained environments. The DO and PH methods are designed to be plug-and-play, facilitating seamless integration into future lightweight models.

Digital watermarks are hidden signatures embedded in media like images or videos to help prove ownership and protect against unauthorized copying. However, most high-performing watermarking tools rely on large models that are too big to run on devices with limited resources.We propose a new watermarking technique that performs well with compact AI models. Instead of directly adapting existing methods, we identified key training challenges for small models and introduced two solutions: one redefines the training objective, and the other introduces a lightweight detachable module used only during training.Our method keeps watermarks invisible and robust, while reducing model size and computational cost. This work makes it easier to apply copyright protection in real-world settings where speed, storage, and power consumption are critical.