🎥 🦭 Video Seal: Open and Efficient Video Watermarking

Official implementation of Video Seal: training and inference code for video watermarking, and state-of-the-art open-sourced models.

This repository includes pre-trained models, training code, inference code, and evaluation tools, all released under the MIT license, as well as baselines of state-of-the-art image watermarking models adapted for video watermarking (including MBRS, CIN, TrustMark, and WAM) allowing for free use, modification, and distribution of the code and models.

[paper] [arXiv] [Colab] [Demo]

Original	Video Seal output	The watermark (normalized for visibility)

Quick start

import torchvision
import videoseal
from videoseal.evals.metrics import bit_accuracy

# Load video and normalize to [0, 1]
video_path = "assets/videos/1.mp4"
video = torchvision.io.read_video(video_path, output_format="TCHW")
video = video.float() / 255.0

# Load the model
model = videoseal.load("videoseal")

# Video Watermarking
outputs = model.embed(video, is_video=True) # this will embed a random msg
video_w = outputs["imgs_w"] # the watermarked video
msgs = outputs["msgs"] # the embedded message

# Extract the watermark message
msg_extracted = model.extract_message(video_w, aggregation="avg", is_video=True)

# VideoSeal can do Image Watermarking
img = video[0:1] # 1 x C x H x W
outputs = model.embed(img, is_video=False)
img_w = outputs["imgs_w"] # the watermarked image
msg_extracted = model.extract_message(imgs_w, aggregation="avg", is_video=False)

Installation

Requirements

Version of Python is 3.10 (pytorch 2.5.1, torchvision 0.20.1, torchaudio 2.5.1, cuda 12.1). Install pytorch:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install -e .

For training, we also recommend using decord:

pip install decord

Note that there are some issues with installing decord: dmlc/decord#213 Everything should be working without decord for inference, but there may be issues for training in this case.

Download the Video Seal model

The video model is automatically downloaded through Hugging Face by doing videoseal.load("videoseal"). If you wish to download the model manually, you can do so by running the following command:

wget https://dl.fbaipublicfiles.com/videoseal/checkpoint.pth

(or click here). You then have to update the checkpoint path in the model card file.

Download the other models used as baselines

We do not own any third-party models, so you have to download them manually. We provide a guide on how to download the models at docs/baselines.md.

VMAF

We provide a guide on how to check and install VMAF at docs/vmaf.md.

Inference

Audio-visual watermarking

inference_av.py

To watermark both audio and video from a video file. It loads the full video in memory, so it is not suitable for long videos.

Streaming embedding and extraction

inference_streaming.py

To watermark a video file in streaming. It loads the video clips by clips, so it is suitable for long videos, even on laptops.

Example:

python inference_streaming.py --input assets/videos/1.mp4 --output_dir outputs/

Will output the watermarked video in outputs/1.mp4 and the binary message in outputs/1.txt.

Full evaluation

videoseal/evals/full.py

To run full evaluation of models and baselines.

Example to evaluate a trained model:

python -m videoseal.evals.full \
    --checkpoint /path/to/videoseal/checkpoint.pth \

or, to run a given baseline:

python -m videoseal.evals.full \
    --checkpoint baseline/wam \

Training

Example run of a simple run on a node of 8 GPUs:

torchrun --nproc_per_node=8  train.py --local_rank 0 --debug_slurm \
  --balanced True --total_gnorm 1.0 --lambda_dec 1.0 --lambda_det 0.0 --lambda_d 0.1 --lambda_i 0.5 \
  --scaling_w 1.0 --scaling_i 1.0 --nbits 96 --perceptual_loss mse --seed 444 \
  --scheduler CosineLRScheduler,lr_min=1e-6,t_initial=1000,warmup_lr_init=1e-8,warmup_t=50 --optimizer AdamW,lr=1e-5 \
  --epochs 1000 --eval_freq 5 --full_eval_freq 5 --batch_size_video_eval 1 --batch_size_video 1 --batch_size_eval 32 --batch_size 16 --iter_per_valid 10 \
  --videoseal_step_size 16 --video_start 800 --prop_img_vid 0.5 --iter_per_epoch 1500 \
  --extractor_model sam_small --embedder_model unet_small2 --augmentation_config configs/augs.yaml \

License

The model is licensed under an MIT license.

Contributing

See contributing and the code of conduct.

Citation

If you find this repository useful, please consider giving a star ⭐ and please cite as:

@article{fernandez2024video,
  title={Video Seal: Open and Efficient Video Watermarking},
  author={Fernandez, Pierre and Elsahar, Hady and Yalniz, I. Zeki and Mourachko, Alexandre},
  journal={arXiv preprint arXiv:2412.09492},
  year={2024}
}

facebookresearch/videoseal