OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Lidar Mos

LiDAR-MOS (LMNet) is Python research code that segments sequential 3D LiDAR scenes into moving and static objects using temporal information and range-image-based segmentation networks.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Lidar Mos — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you need a research implementation for temporal moving-versus-static segmentation in 3D LiDAR, or want to investigate dynamic-object filtering ahead of LiDAR odometry, SLAM or mapping.

Who it is for

  • Researchers studying 3D LiDAR motion detection and segmentation.
  • Robotics developers exploring dynamic-object handling for LiDAR odometry, SLAM or mapping.

Documented uses

  • Separate moving objects, such as moving cars, from static objects such as parked cars in sequential LiDAR data.
  • Generate predictions using the documented pretrained SalsaNext model and toy dataset.
  • Train a SalsaNext-based LiDAR-MOS network using KITTI Odometry and SemanticKITTI data.
  • Evaluate LiDAR moving-object segmentation against the project's SemanticKITTI-based benchmark.
Before you choose

Requirements & limits

The repository is public and released under the MIT license.

  • The approach uses range images and modifies the data loader and network input to exploit sequential information.
  • The README documents baseline paths for SalsaNext and RangeNet++, and identifies MINet among the implementations used as a basis.
  • The documented SalsaNext setup specifies PyTorch 1.5.1+cu101.
  • Whole-dataset inference requires KITTI Odometry data and adjusted paths; training from scratch requires both KITTI Odometry and SemanticKITTI datasets.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run the documented SalsaNext toy-dataset inference

Generate LiDAR-MOS predictions for the toy dataset's validation split using the supplied pretrained SalsaNext model.

Before you start

  • The repository's toy dataset, decompressed in the data folder following data/README.md.
  • The pretrained model archive salsanext_residual_1, unzipped before use.
  • SalsaNext baseline setup and the repository's documented PyTorch 1.5.1+cu101 version.

Steps

  1. Download and decompress the toy dataset into the data folder, following data/README.md.

  2. Download and unzip the pretrained SalsaNext model in the data folder.

  3. From the repository, change to mos_SalsaNext/train/tasks/semantic.

  4. Run python3 infer.py -d ../../../../data -m ../../../../data/model_salsanext_residual_1 -l ../../../../data/predictions_salsanext_residual_1_new -s valid.

Expected outcome

The inference command is configured to generate LiDAR-MOS predictions for the toy dataset's validation split and write them to the specified predictions path.

Guide limitations

  • This recipe covers the repository's toy-data inference example, not whole-dataset evaluation.
  • The README directs users to the SalsaNext README for its setup; this guide does not establish machine-specific compatibility.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
7 November 2026
Original website title
PRBonn/LiDAR-MOS · Official homepage title · retained in its source language
How we review sourcesSuggest a correction