Deploying SpatialLM requires configuring a specialized deep learning environment, which consists of three main levels of preparation:
- base runtime: Python 3.11 and PyTorch 2.4.1 frameworks, it is recommended to create an isolated environment via Conda to avoid version conflicts.
- GPU acceleration support: CUDA 12.4 toolkit must be installed and NVIDIA driver compatibility must be verified. Model inference requires a graphics card with at least 8GB of video memory.
- version-specific component::
- Version 1.0 requires compilation of the torchsparse library (a key component for processing sparse point clouds)
- Version 1.1 requires the flash-attn optimization package to accelerate attention calculations
It is worth noting that the project uses poetry to manage dependencies, and the installation needs to be performed first by executing thepoetry installLoad the base package, then select additional components by version. The full configuration takes up about 15GB of disk space.
This answer comes from the articleSpatialLM: Sweep the room, AI automatically draws 3D models for youThe
May not be reproduced without permission:AI productivity tools " What environment dependencies do I need to prepare for 3D scene reconstruction using SpatialLM?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil