Differentiable point clouds in pytorch

Pytorch implementation of paper https://arxiv.org/abs/1810.09381

To install dependencies use conda.

Install with:

conda create -n point_clouds python=3.7
conda activate point_clouds
conda install pytorch torchvision scipy matplotlib open3d=0.7 tqdm jupyter -c pytorch -c open3d-admin
jupyter notebook

Then use point_clouds.ipynb to train and evaluate model. Resulting point cloud trained without camera supervision: https://youtu.be/_8aONZdoBzQ