This repo contains the implementation of the paper "Eva: Efficient Privacy-Preserving Proof of Authenticity for Lossily Encoded Videos" (IEEE S&P 2025). The dataset can be found on Hugging Face.
The implementation is based on Sonobe, a library for folding-based IVC. We forked Sonobe and built a variant of Nova with support for lookup arguments in the folding-schemes folder. The code for video authentication can be found in video.
Hardware requirements:
- 64 GB of RAM.
- An Nvidia GPU. We used an RTX 3080 with 12 GB of VRAM, but 8 GB or even 6 GB should also work.
- 50 GB of free disk space.
You need to have packages for C/C++ development (build-essential for apt, development-tools for dnf, base-devel for pacman), Rust, and CUDA installed on our machine.
Before running the examples, download the data_parsed folder from Hugging Face and set the environment variable DATA_PATH=/<path>/<to>/data_parsed.
You can also use your own video, but you need to extract the original & prediction macroblocks as well as the quantized coefficients from the JM library. Instructions for dataset preparation will be provided in the future.
Simply execute cargo run --release --example=<example>.