This repo holds the code for the paper: Deep-PCAC: An End-to-End Deep Lossy Compression Framework for Point Cloud Attributes https://ieeexplore.ieee.org/document/9447226
tensorflow-gpu 1.13.1
open3d
nvidia-docker2
cuda10.0
python 3.6
We trained four models with four different latent_points (latent_points=256,200,128,64). You can modify the latent_points to change the rate.
Pre-trained models are stored in /model/256, /model/200, /model/128, /model/64. Please modify the "checkpoint" file in these folders and change the absolute path to find the ckpt. (I have midify the path of "checkpoint" file, you can directly run them. If you have some problems about the path, please modify the path in "checkpoint" file)
I have built a docker to aviod setting up the environment. You can pull the docker frome dockerhub.
https://hub.docker.com/r/xhsheng/deep_pcac
docker pull xhsheng/deep_pcac
Then you need to run the nvidia-docker and mount the local directory using the following command:
sudo nvidia-docker run -v /home/Deep-PCAC-main/:/media/deep-pcac -it xhsheng/deep_pcac:latest
After entering the container, you can run the code using the following commands.
python mycodec.py compress --input="./testdata/soldier_vox10_0690.ply" --ckpt_dir='./model/256/' --latent_points=256
python mycodec.py decompress --input="./testdata/soldier_vox10_0690" --ckpt_dir='./model/256/' --latent_points=256
I am sorry that I have no time to sort my code becuase I am being an intern. Therefore, I can only release a .so file of my model. If you have any questions, please contact me (xhsheng@mail.ustc.edu.cn). I will try my best to solve your concerns.