Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

LD-RoViS: Training-free Robust Video Steganography for Deterministic Latent Diffusion Model

Authors: Xiangkun Wang, Kejiang Chen, Lincong Li, Weiming Zhang, Nenghai Yu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to evaluate LD-Ro Vi S, demonstrating its superior performance over existing methods in terms of capacity, robustness, and security. Our implementation is available at https://github.com/xiangkun1999/LD-Ro Vi S.
Researcher Affiliation Academia Xiangkun Wang1,2 Kejiang Chen1,2 Lincong Li1,2 Weiming Zhang1,2 Nenghai Yu1,2 1University of Science and Technology of China, China 2Anhui Province Key Laboratory of Digital Security, China EMAIL EMAIL
Pseudocode Yes The embedding and extraction procedures are shown in Algorithm 1 and Algorithm 2, respectively.
Open Source Code Yes Our implementation is available at https://github.com/xiangkun1999/LD-Ro Vi S.
Open Datasets Yes For evaluation, we use Vid Pro M [47], a large-scale and diverse text-to-video prompt dataset.
Dataset Splits Yes Using ffmpeg, we decode these videos to obtain 8,100 pairs of cover and stego frames, with 4,000 used for training, 600 for validation, and 3,500 for testing.
Hardware Specification Yes All subsequent experiments are conducted on these 100 prompts with seed = 99, ks = 5.0 and run on four NVIDIA RTX A6000 GPUs, each with 48 GB of VRAM.
Software Dependencies No The paper mentions software like 'FFmpeg' and 'Python s time package' and specific model versions like 'LTX-Video [48] (ltxv-2b-0.9.8-distilled version)' but does not provide specific version numbers for general software dependencies such as programming languages, deep learning frameworks (e.g., PyTorch, TensorFlow), or CUDA versions.
Experiment Setup Yes In our experiments, we set the hyperparameters as τ1 = 0.32, τ2 = 0.02, and cfg = 16. Additional experiments and analysis of these hyperparameters can be found in the Appendix A, including experiments on a non-deterministic model (LTX-Video [48]).