prs-eth/LitePT

[CVPR 2026 Highlight] LitePT: Lighter Yet Stronger Point Transformer

What it solves

LitePT addresses the high computational cost and memory requirements of 3D point cloud architectures, specifically targeting the inefficiencies of Point Transformer V3 (PTv3). It provides a high-performance alternative that is significantly lighter and faster without sacrificing accuracy.

How it works

LitePT is a lightweight 3D point cloud architecture available in multiple scales (Small, Base, Large). It utilizes a specialized component called PointROPE, which is provided in both a CUDA-optimized version for maximum performance and a pure PyTorch implementation for broader compatibility. The architecture is designed to be more efficient in terms of parameters and memory footprint while maintaining or exceeding the performance of state-of-the-art backbones.

Who it’s for

This project is for researchers and developers working with 3D computer vision, specifically those focusing on semantic segmentation, instance segmentation, and object detection within 3D point clouds.

Highlights

  • High Efficiency: The LitePT-S variant has 3.6x fewer parameters, 2x faster runtime, and 2x lower memory usage compared to PTv3.
  • Scalable Performance: Even the large variant (LitePT-L) remains more memory-efficient than PTv3 despite having more parameters.
  • Versatile Applications: Supports semantic segmentation, instance segmentation, and object detection across benchmarks like NuScenes, Waymo, ScanNet, and Structured3D.
  • Flexible Integration: Provides a standalone implementation and a demo script for easy integration into external projects.

Related

  • Project
    facebookresearch/actionmesh

    ActionMesh is an open‑source model that turns short videos into animated 3‑D meshes (or animates an existing mesh). It runs on a GPU, offers a low‑RAM mode for Colab, and ships with a new benchmark (ActionBench).

  • Project
    AirsupHQ/machine-skill

    Machine Skill is a Claude Code Agent Skill that turns a one‑line description of a real machine (e.g., “a turbofan jet engine”) into a photoreal, interactive three.js web page with accurate geometry, material rendering, and a lightweight physics solver. The page can be explored, exploded, cut‑away, or “broken”, and a deterministic 30‑second film can be recorded with a single command. Install via `npx skills add AirsupHQ/machine-skill`, then ask Claude `/machine …` to get an HTML file. The repo contains the skill manifest, a single runtime kit (no build step), physics solver recipes, example machines, and scripts for screenshots and video capture. It’s aimed at educators, creators, and engineers who need quick, physics‑backed visualisations of engines, pumps, compressors, watches, etc., but it uses simplified models (not CFD/FEA) and requires a modern GPU for the heaviest scenes.

  • Project
    phongdaot/MocapAnything

    An end-to-end motion capture system that converts monocular videos into 3D joint rotations for arbitrary skeletons, including animals and custom rigs.

  • Project
    LTplus-AG/ifc-lite

    IFClite is an open‑source, Rust‑powered toolkit (exposed via npm packages, a CLI and a WebGPU viewer) for parsing, querying, editing, validating, rendering and exporting IFC building‑information models. It supports IFC2X3, IFC4/4X3 and IFC5, offers exact‑arithmetic geometry, streaming first‑render, and integrations for Three.js, Babylon.js, Python, and AI agents.