InstantSplamp: Fast and Generalizable Stenography Framework for Generative Gaussian Splatting

a) Traditional 3D generation requires separate steps for object generation and watermarking, leading to significant time overhead. • (b) Our method, InstantSplamp, unifies 3D generation and watermarking into a single process, maintaining the generation time and reducing watermarking overhead to near-zero, significantly improving efficiency. ➢Experimental Results Results of Watermarking Indomain Objects ➢Framework * During (a) Hidden Information Embedding, InstantSplamp incorporates the DINOv2 features of the hidden information into the intermediate feature of Gaussian generation via cross-attention. * In (b) Hidden Information Recovery, a U-Net-based decoder is employed to retrieve the hidden information from the rendered image under the checking pose. * Through the optimization process, (c) Adaptive Gradient Harmonization is used to maintain a balance between the rendering and hidden recovery.