Product Overview

ℹ️

Data Privacy Statement

The Lixel CyberColor Studio (Linux Server) is designed with data privacy as a fundamental principle, ensuring that sensitive information remains securely within the customer’s private environment. The software operates exclusively within a customer-controlled Docker environment, and all interactions with the business system are limited to read-only operations, authorized by the customer, ensuring no write access or data modification occurs. Key data privacy measures include:

1. Isolated Processing: The LCC software processes data entirely within the customer’s private infrastructure with no data transmission to external servers or third-party services.
2. Authorization-Driven Access: All API calls require authentication via a customer-provided access key, ensuring only authorized interactions.
3. No External Data Reading: The software strictly operates on data provided within the specified working directory. It does not access, read, or transmit data from the private server or any external systems.
4. Customer Control: All input and output directories are defined by the customer, giving them complete control over the data flow and ensuring compliance with internal data governance policies.

Introduction

LCC Studio (Linux Server) is a high-performance, scalable containerized solution designed for efficient 3D reconstruction. It supports multiple reconstruction modes, including single-map reconstruction, map fusion, and aerial-ground map fusion, while leveraging cluster computing to handle large-scale reconstruction tasks. Built on Docker, LCC seamlessly integrates with client enterprise system, offering flexible deployment options. Byutilizing multi-GPU parallel computing, it significantly enhances reconstruction performance. The software also enables data export for secondary development and ecosystem expansion.

Overview

Note: Single-map reconstruction, map fusion, aerial-ground map fusion, developer data export, and cluster mode are advanced features (Paid Add-ons). Feature availability depends on your specific requirements. Contact the XGRIDS sales team for details.

• Multi-Mode Reconstruction: Supports single-map reconstruction, map fusion, and aerial-ground map fusion. When following standardized procedures, this version works best up to 300 minutes of ground scanning and 15,000 drone images for 3DGS reconstruction.
• High-Performance Computing: Multi-GPU parallel acceleration is supported only for map fusion and aerial-ground map fusion, enabling efficient processing of large-scale datasets.
• Data Accessibility: Supports developer data export, enabling secondary development and ecosystem expansion.
• System Compatibility: Provides an HTTP API for seamless integration with other business systems, facilitating automated processing and application extensions.
• Algorithm Upgrade: This version features algorithm optimizations that improve memory usage and detail rendering.
• Authorization Methods: This version supports online authorization, offline hardware authorization, and offline software authorization.

Workflow

Note: Upon startup, LCC only retrieves authentication information from the XGRIDS server. This authorization verification ensures secure system access. While this requires server interaction, no work data is involved—your private data is never accessed, transmitted, or processed during this step.

Private DeploymentCustomer Linux Server (Nvidia Docker Runtime + Nvidia Driver)LCC DockerContainer 0(Main Node)GPU 0LCC DockerContainer 1(Sub Node 1)GPU 1LCC DockerContainer 2(Sub Node 2)GPU 2ServiceRegistryDockerContainerData StoragemountCustomerBusinessSystemAPICustomer UserUIUSB DongleorOnlineAuthorizationGet authorizationNotes: LCC retrieves authorizationinfo at startup; no work datais transmitted.

Architecture

Business
Layer

Industry Applications

Permission Control

API Service Bus

Software

Service

Capability
Layer

Spatial Reconstruction

Multi-SLAM

Aerial Triangulation

Panoramic Reconstruction

Multi-sensor Fusion

Spatial Understanding

Heterogeneous Computing

Gaussian Splatting

Neural Rendering

Multi-source Data Fusion

Multi-map Fusion

Air-ground Fusion

High-performance Distributed Computing

Large-scene Optimization

Data Compression

LOD Level Management

Streaming Loading

Multi-platform Roaming

XGRIDS
Distributed
Computing
Framework

Distributed Task Scheduling

Data Validation and Packaging

Task Creation and Chunking

Task Status Management

Task Data Management

Distributed Resource Management

Resource Allocation

Resource Status Management

Resource Registry

Virtualized
Resource Layer

Distributed Computing Power Virtualization

Network Interface Layer

Transport Layer

Computing Resource Encapsulation

Storage Layer

Distributed Storage Virtualization

Local Storage Management

Network Storage Management

Containerization Management

Infra-
structure

GPU

CPU

FGPA

ASICs

Storage

Network

Container

Integration

Prepare Data

• Organize input data in the designated working directory as per the documentation requirements.

Initiate Multi-Map Fusion Reconstruction

• Start the reconstruction task using the provided HTTP API.

Monitor Progress

• Poll the API for real-time status updates until the task is complete.

System Requirements

Hardware Specifications

LCC reconstruction performance primarily depends on CUDA Cores and FP32 TFLOPS (floating point operations per second). Factors such as NVLink, Tensor Cores (AI computing), FP64, and memory size (≥32GB) have a relatively minor impact. As a result, certain high-end AI GPUs (e.g., A100, H100) may underperform compared to lower-tier models with higher CUDA Core counts and FP32 computing power (e.g., A10, L40).

• LCC Reconstruction Workload Distribution:

  • GPU-intensive algorithms: ~50% of workload, significantly impacted by GPU performance, especially CUDA Core count and FP32 computing power.
  • CPU-intensive algorithms: ~50% of workload, significantly impacted by CPU frequency and physical cores.

Recommended GPUs

• When selecting a GPU, it is recommended to choose a model based on task requirements and budget. Below is a ranked list of recommended GPUs based on performance:

  GPU Model	Performance Ranking	Key Advantages	Best For
  L40	1 (Best)	Superior CUDA Core count and FP32 computing	High-volume professional work
  RTX 4090	2	Excellent CUDA Core performance with efficient FP32 computing	Most professional reconstruction tasks
  RTX 4090 D	3	A domestic version of the RTX 4090, offering strong CUDA Core performance and efficient FP32 computing	High-performance reconstruction with cost considerations
  L20	4	Strong CUDA Core and FP32 performance	Budget-conscious professional use
  RTX 3090	5	Great price-to-performance value	Cost-effective professional work
  A10	6	Budget-friendly option	Basic reconstruction tasks

• Notes:

  • Memory Requirements: 24GB VRAM is sufficient for most tasks. Increasing beyond 32GB offers minimal performance benefits unless processing extremely large scenes.
• A GPU performance reference table is provided below. Models marked in red with an asterisk (*) have been benchmarked, while other models are estimated based on similar architectures.

Multi-GPU Configuration

• When reconstructing large-scale scenes, a multi-GPU setup can provide significant acceleration. The relationship between scene size (ground scanning duration + number of drone images) and the maximum number of usable GPUs is shown in the graph below.