Using Laplacian GAN for 3D pointcloud
MIT License
Official Pytorch repository for VISAPP 2020 Latent-Space Laplacian Pyramids for Adversarial Representation Learning with 3D Point Clouds
This repository is devoted to the topic of three-dimensional point clouds generation. It is proposed to implement the architecture of a multilevel generative adversarial neural network. This work is based on the concept set forth in the article Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks, whose authors propose the structure of a multilevel generative adversarial neural network, for obtaining a better quality in high resolution images, and on the article Representation Learning and Adversarial Generation of 3D Point Clouds, showing the efficiency of point cloud generation by using the auto-encoder and generative adversarial network, which generates a latent representation, instead of direct generation of a point cloud. You should check he's repository https://github.com/optas/latent_3d_points ,because this one based on it. Code that provided from optas is based on tensorflow 1.3, and my code on Pytorch 0.3 +
Take a look at usage here https://github.com/optas/latent_3d_points . After you come back just take a look at new files in folders notebooks/ and src/
Build the docker container:
docker build -f Dockerfile --tag artonson/3dlapgan:latest .
Then enter the container by running it (don't forget to name the containers accordingly and remove them post-usage!):
docker run --rm -it --name 3ddl.artonson.0.3dlapgan --runtime=nvidia -v /home/artonson/repos/ThreeDLAPGAN:/code -p 3340:3340 artonson/3dlapgan:latest
Remember our container naming conventions: 3ddl.<username>.<gpu-ids-list>.<customsuffix>
.
After entering the container shell, you will be able to run the Jupyter notebook:
cd /code
jupyter notebook --NotebookApp.token=abcd --ip=0.0.0.0 --port 3340 --allow-root
and your token will be abcd
.
First level(512)
Second level(1024)
Third level(2048)