Skip to content

Latest commit

 

History

History
71 lines (46 loc) · 1.75 KB

README.md

File metadata and controls

71 lines (46 loc) · 1.75 KB
Raw

Docker for $D^2$SLAM

Our Docker image includes:

  • ros-noetic
  • ceres-2.1.0
  • onnxruntime-gpu-1.12.1
  • libtorch-latest
  • LCM
  • faiss
  • OpenCV4 with CUDA
  • OpenGV
  • Backward
  • $D^2$SLAM

Docker PC

To build the Docker image for PC, run the following command:

$ make amd64

Docker for Jetson

This Docker file can be built on a MacBook with Apple Silicon (M1 or M2), X86_64 PC with qemu support or on Jetson. However, in our tests, building on Jetson is takes hours and building on Qemu is even more slow.

We highly recommend building the image on a MacBook Pro with M1/M2 Max. This is possibly the fastest way.

To build the Docker image for $D^2$ SLAM, run:

$ make jetson

No-CUDA configuration

Target arm64 (Dockerfile.arm64_ros1_noetic) and x86_no_cuda (Dockerfile.x86_no_cuda) provide non-cuda configuration for arm64 and X86-64 devices. Others will depends on CUDA. D2VINS only has fully abaility when using CUDA, features will be unsupported without CUDA:

  • Superpoint and NetVLAD. You can only work with LK optical tracking.
  • Loop clousure and pose graph
  • Multi-robot localization will be disabled without CUDA
  • Depth generation

Basically, without CUDA, D2SLAM will become a mono/stereo/quad camera visual-inertial odometry (VIO).

Build Base Container (Optional)

To build the base image for $D^2$SLAM (which contains the environment for those who would like to modify it), run:

$ make jetson_base

Then, in Dockerfile_jetson_D2SLAM, change:

FROM xuhao1/d2slam:jetson_base_35.1.0

to your own image name:

FROM your-image-name/d2slam:jetson_base_35.1.0

This Docker image has been tested on Jetpack 5.0.2/35.1.0 with Xavier NX.