3d-perception-dockerfiles

---

This repository contains Dockerfiles for the Open3D and will be updated regularly with updated versions to include various other libraries required to carry out 3D-perception.

Currently, the repo has two Dockerfiles under the docker folder and both of them are tested with a base example is testfor an x86_64:

Dockerfile :

System	Version
Ubuntu	20.04
Docker	20.10.14

with :

System	Version
pip	20.3
Python	3.8
Open3D	0.15.2

System Prerequisites:

• Docker

The PyPI has open3d with version no. 0.15.2 instead of 0.15.1 . I believe it is just a new commit fully resolved with any issues from 0.15.1

Dockerfile.pt3d :

1. Has PyTorch cuda Ops built for Open3D
2. Open3D built from source

System	Version
CUDA	11.6
Open3D	0.15.1
pip	20.3
Python	3.8
Pytorch	1.13

System Prerequisites:

• Docker
• >=22 GB storage memory
• Nvidia Drivers>=515
• Nvidia runtime container toolkit installed on host

---

Steps to use the Docker files to create Docker Images:

Clone this repository:

~$ git clone https://github.com/raghavduddala/3d-perception-dockerfiles.git
~$ cd 3d-obstacle-detection/

Edit the Makefile with the correct DOCKERFILE for the required docker image you want to built. for example for building the Open3D-PyTorch container, in Makefile line 4

DOCKERFILE := Dockerfile.pt3d

and use the Makefile commands to building, launching and extending terminals to the containers In the first terminal, to build and launch the Open3D Docker Container

~$ make build
~$ make dev

or the command below to launch the cuda-related containers.

make cuda-dev 

In the Second terminal, to attach a terminal to the container

~$ make shell

After the docker container is launched, run the basic test file from Open3D tutorials here to check if the installation and build is fine.

~$ cd src/
~$ python3 test_open3d.py 

A sample .ply file containing a Point cloud from Open 3D is downloaded and visualized in the following way.

---

Usage examples when using with other repos that require the docker container

My Repos using this include https://github.com/raghavduddala/point_clouds_open3d.git

where the repo we want to run can be cloned under the src directory of this docker workspace folder.

---

Please support by giving a star if you find this repo helpful for building the current containers with the dockerfiles or in building your custom docker images