About this guide#
COPYRIGHT©2025 Vueron Technology Co., Ltd. All rights reserved.
This guide is provided as a reference and training aid for VueX purchased by customers of Vueron Technology. It may be distributed only to the customer’s employees and agents for reference and internal training purposes. All information and content within this document are the property of Vueron Technology, and unauthorized reproduction is strictly prohibited without written permission from Vueron Technology. Furthermore, this content may not be processed or excerpted without permission. Vueron Technology makes no express or implied warranties regarding the content of this document, and the information provided is subject to change without notice. Revisions may be issued to reflect such changes. The software described in this guide is provided under a license or non-disclosure agreement. Except where explicitly permitted by the agreement, reverse engineering or similar activities are prohibited.Getting Started with VueX#
To get started with VueX, an AWS account is required. You can either click "Get started" on the www.vueron.ai page to log in to your AWS account, or search for VueX directly on AWS Marketplace and click the Subscription button to start using VueX.
By following this process, users will proceed with the VueX sign-up application. Once the sign-up is complete, the account manager will contact the provided email to share product information, contract details, and instructions on how to use VueX.Collection#
To start with VueX, you need to upload the acquired LiDAR data. VueX supports the PCD format (*.pcd). Users can create a folder by clicking "New folder" to upload their data, then either click "Upload" within the created folder or use Drag & Drop to upload the data.Annotation#
To train an AI model, the uploaded data needs to be processed. VueX uses its powerful AI capabilities to quickly and automatically label the uploaded data.(1) Create Project#
Click "Create project" to create a project. Project Name and Member preset are required fields, and you can choose the desired Object class for annotation.(2) Create Task#
Select the created project and click "Create Task" to choose the uploaded data from the Collection and assign it to the task.Task Folder: The folder specified in the Collection will be loaded. If needed, you can change the folder by selecting "Select Folder."
Task Name: The folder name is the default, but it can be changed.
Start/Due Date: You can specify the task schedule.
Frame interval: Refers to the interval between the frames you wish to use.
Target Object: The object class selected when creating the project will be set as the default and can be changed.
Confidence: Sets a threshold, and predictions with confidence below this threshold will be excluded from the AI-annotation results. You can set the Confidence for each Object Class.
Distance: The value that represents the distance of the AI-annotation result relative to the center, forming a circular shape.
(3) Annotation#
There are three ways to perform annotation in VueX.1. AI-annotation#
AI-annotation for individual frameIn the created task, select an individual frame and click the AI button at the top-left to perform AI-annotation for that frame. For accurate AI-annotation, you need to select the appropriate model for the task through Select Model. Confidence and Distance values are set during project creation, and Available shows the number of frames that can be annotated.
Available models and their scenarios are as follows:
| Model | Environment | Updated |
|---|
| L_BiKal_mobility | Moving objects like cars and robots | 2025-05 |
| L_BiKal_static_outdoor | Fixed outdoor infrastructure like intersections and roads | 2025-05 |
After performing AI-annotation, users can manually edit the results if necessary.
Click the "AI Run" button to perform AI-annotation for all frames in the task. To ensure accurate AI-annotation, you must select the appropriate model for the task through Select Model. Confidence and Distance values are set during project creation, and Available shows the number of frames that can be annotated.
Available models and their scenarios are as follows:
| Model | Environment | Updated |
|---|
| L_BiKal_mobility | Moving objects like cars and robots | 2025-05 |
| L_BiKal_static_outdoor | Fixed outdoor infrastructure like intersections and roads | 2025-05 |
※ The "AI Run" feature applies to all frames in the task and cannot be undone. Therefore, it is recommended to first test AI-annotation on a few frames before applying it to all frames.After AI-annotation, corrections can still be made manually.
2. AI-assistance#
AI-assistance helps by creating a bounding box based on clustered points to assist with annotation. After identifying the clustered points, activating the AI-assistance feature will create a 3D bounding box that fits the object. After creating the box, you can select the desired class and save it.3. Manual Annotation#
Users can manually perform annotation by working directly on the data.(4) Creating Annotation Boxes#
To generate box data, a Verify process is required. Verification can be done frame-by-frame, or you can use the "All Verify" button to verify all frames at once. Once the verification is completed, the box data will be available for download.Modeling#
VueX provides powerful pre-set LiDAR perception AI models. Users can acquire data based on the provided Baseline models and create models optimized for their specific environment. These models are designed to work not on cloud or high-performance servers, but on Edge devices.(1) Model Training#
Model training is performed by training data created through annotation using a baseline model. To begin training, you must select the data you want to use from the annotated datasets and divide it into Training dataset and Testing dataset. There are two methods for splitting the datasets:Auto split : Click "Select data" to choose datasets from the completed annotations, and the system will automatically split them into Training dataset and Testing dataset.
User split : Click "Select data" to choose datasets from the completed annotations, and you can define the ratio for Training dataset and Testing dataset for each dataset.
Available baseline models and their scenarios are as follows:
| Model | Scenarios | Updated |
|---|
| Mobility_Baseline | Moving objects like cars and robots | 2025-05 |
| Static-outdoor_Baseline | outdoor infrastructure like intersections and roads | 2025-05 |
Once model training is completed, the trained model based on the Baseline model will be created. The Performance analysis for this model includes quantitative evaluation results such as AP, Precision-Recall, F1 Score, PR Curve, Recall/Confidence Score, and Precision/Confidence Score, as well as qualitative evaluation results displayed through the Viewer for the perception results.Deployment#
Deployment refers to the process of deploying the trained model to Edge Devices. The models provided by VueX are available in an Engine file format, designed to operate on Edge Devices. The supported Edge Devices are as follows:| Controller | Updated |
|---|
| Nvidia AGX Orin based Embedded pc | 2025-05 |
| Nvidia Orin NX based Embedded pc | 2025-05 |
After training is complete, select the model you want to deploy and click the "Deployment" button to begin the deployment process. Once the deployment is complete, the model will be created in the Collection, and it will be available for download.The downloaded model can be easily used on any Edge Device with the help of the VueX Inference API Manual.Modified at 2025-08-07 14:04:10
