Deep AutoEncoder-based Lossy Geometry Compression for Point Clouds

Created by Wei Yan

Introduction

We propose a general autoencoder based architecture for lossy geometry point cloud compression.

Dependencies

Requirements:

Python 2.7+ with Numpy, Scipy and Matplotlib
Tensorflow (version 1.10+)
TFLearn

Our code has been tested with Python 2.7, TensorFlow 1.10.0, TFLearn 0.3.2, CUDA 9.0 and cuDNN 7.0 on Ubuntu 14.04.

Installation

To be able to train your own model you need first to compile the EMD/Chamfer losses. In latent_3d_points_entropy/external/structural_losses we have included the cuda implementations of Fan et. al.

cd latent_3d_points_entropy/external

with your editor modify the first three lines of the makefile to point to 
your nvcc, cudalib and tensorflow library.

make

Data Set

We provide ~57K point-clouds, each sampled from a mesh model of ShapeNetCore with (area) uniform sampling. To download them (1.4GB):

cd latent_3d_points_entropy/
./download_data.sh

The point-clouds will be stored in latent_3d_points_entropy/data/shape_net_core_uniform_samples_2048

Use the function snc_category_to_synth_id, defined in src/in_out/, to map a class name such as "chair" to its synthetic_id: "03001627". Point-clouds of models of the same class are stored under a commonly named folder.

Usage

To train a point-cloud encoder and decoder look at:

cd notebooks
python train_single_class_ae.py

You can change the number of points and the category of training set at the top lines.

To compress a point cloud from shapenet:

cd notebooks
python compress.py --input_pointcloud_file \
 /YOUR/POINTCLOUD/FILE --output_dir /OUTPUT/DIR \
  --model_dir /PRETRAINED/MODEL/DIR  \ 
  --number_points point cloud number of farthest point sample. It must in accordance with model type!

For example:

python compress.py --input_pointcloud_file /home/yw/Desktop/latent_3d_points_entropy/data/shape_net_core_uniform_samples_2048/03001627/1a6f615e8b1b5ae4dbbc9440457e303e.ply --output_dir ./compressed_data --model_dir /home/yw/Desktop/latent_3d_points_entropy/data/chair_model/chair_inout_point1024 --number_points 1024

We provide four classes pretrained model (car,chair,table,airplane) in ./data.

To decompress a point cloud from your latent code:

cd notebooks 
python decompress.py --latent_code_dir /YOUR/LATENT/CODE/DIR --decompress_dir /DECOMPRESS/DIR --model_dir /MODEL/DIR