INR-PCC

Source code for paper "Point Cloud Compression with Implicit Neural Representations: A Unified Framework".

Requirements

• pytorch 1.9.0
• open3d 0.18.0
• deepCABAC 0.1.0: https://github.com/fraunhoferhhi/DeepCABAC
• tmc3 v23.0-rc2: https://github.com/MPEGGroup/mpeg-pcc-tmc13
• pc_error: https://github.com/minhkstn/mpeg-pcc-dmetric
• 8iVFB dataset: https://plenodb.jpeg.org/pc/8ilabs

Usage

The two softwares tmc3 and pc_error are already contained in this repo. Use the following command to change file permissions.

sudo chmod 777 tmc3 pc_error

Geometry compression

Make sure that the configuration file config.yaml is available in exp/geometry. Then use the following commands for geometry compression.

python train.py --cloud_path=loot_vox10_1200.ply --exp_dir=exp/geometry
python test.py --cloud_path=loot_vox10_1200.ply --exp_dir=exp/geometry

The results can be found in exp/geometry/result.csv.

Attribute compression

Make sure that the configuration file config.yaml is available in exp/attribute. Then use the following commands for attribute compression.

python train.py --cloud_path=loot_vox10_1200.ply --geometry_path=exp/geometry/cloud.ply --exp_dir=exp/attribute
python test.py --cloud_path=loot_vox10_1200.ply --geometry_path=exp/geometry/cloud.ply --exp_dir=exp/attribute

The results can be found in exp/attribute/result.csv.

Baseline (G-PCC)

This code allows compression using G-PCC, by directly calling the tmc3 software with predefined configurations.

Use the following command for geometry compression by G-PCC.

python baseline.py --cloud_path=loot_vox10_1200.ply --result_dir=gpcc --pqs=0.75

Use the following command for joint geometry and attribute compression by G-PCC.

python baseline.py --cloud_path=loot_vox10_1200.ply --result_dir=gpcc --encode_colors --pqs=0.75 --qp=34

The results can be found in gpcc/result.csv.