#Copyright Protection of Neural Radiation Fields Using Invertible Neural Network Watermarking Stega4NeRF
This is the official code for "Copyright Protection of Neural Radiation Fields Using Invertible Neural Network Watermarking"
This code requires Python 3. 8
To train a INN:
python train-INN.py
After training for 300 iterations (~8 hours on a single 2080 Ti), you can find the following model at model.pt
.
The training image for NeRF will be used for watermark embedding using INN
python INNtest,encoder.py
---
To train a low-res `lego` NeRF:
python run_nerf.py --config configs/lego.txt
After training for 50k iterations (~3 hours on a single 2080 Ti)
---
The rendered image is processed using the Image Quality Enhancement module.
---
python 恢复网络.py
---
The processed image is then used to extract the watermark using INN
---
python INNtest,decoder.py
---
# Acknowledgements
[NeRF](https://paperswithcode.com/paper/nerf-representing-scenes-as-neural-radiance#code) models are used to implement Copyright Protection of Neural Radiation Fields Using Invertible Neural Network Watermarking.