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

Pointr

PoinTr is a Python research codebase implementing a geometry-aware transformer model that completes incomplete 3D point clouds. It represents groups of points as position-embedded proxies and uses a transformer encoder-decoder to generate missing geometry. The repository provides pretrained models and workflows for inference, evaluation and training.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Pointr — 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 PoinTr when you have incomplete 3D point-cloud data and want a research implementation for transformer-based completion, or when you need to evaluate or train models within the documented benchmark workflows.

Who it is for

  • Researchers studying 3D point-cloud completion and transformer methods.
  • Computer-vision developers who need to generate completions from point-cloud samples using pretrained weights.
  • Teams evaluating completion models on datasets such as PCN, ShapeNet or KITTI.

Documented uses

  • Generate completed point clouds from incomplete input samples.
  • Save inference outputs and optional visualization images.
  • Evaluate pretrained checkpoints with dataset-specific configurations.
  • Train a point-cloud completion model using the repository’s single- or multi-GPU scripts.
Before you choose

Requirements & limits

The repository is public and MIT-licensed. The documented workflows run from the codebase and require a configured Python, PyTorch and CUDA environment.

  • Inference requires both a configuration file and a compatible pretrained checkpoint.
  • The repository documents pretrained weights for ShapeNet-55, ShapeNet-34, PCN and KITTI.
  • The runtime requirements include CUDA and compiled or installed extensions for Chamfer distance, PointNet++ and GPU kNN.
  • The codebase includes separate scripts for inference, evaluation and training.

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 inference on point-cloud samples

Use a pretrained checkpoint and configuration to generate completed point clouds from an input file or directory.

Before you start

  • Python 3.7 or later, PyTorch 1.7 or later, CUDA 9.0 or later and GCC 4.9 or later.
  • Install the listed dependencies, including torchvision, timm, Open3D and tensorboardX, and build or install the required extensions.
  • A PoinTr configuration file, pretrained checkpoint and input point-cloud file or directory.

Steps

  1. Install the Python dependencies with the command below.

    Documented command
    pip install -r requirements.txt

  2. Build the Chamfer Distance extension with the command below; install the documented PointNet++ and GPU kNN extensions.

    Documented command
    bash install.sh

  3. Run the command below with the configuration and checkpoint paths, plus --pc or --pc_root .

    Documented command
    python tools/inference.py

  4. Optionally use --save_vis_img and --out_pc_root to request visualization images and an output directory.

Expected outcome

The inference script processes the selected point-cloud sample or samples using the chosen configuration and checkpoint; the documented options allow saving visualizations and output point clouds.

Guide limitations

  • The guide depends on having suitable configuration, checkpoint and point-cloud files.
  • The repository notes that extension installation can encounter errors and provides troubleshooting for a missing gridding module.
  • This is repository-based inference, not a hosted service.
Documentation 1 ↗
Source-based practical guide

Evaluate a pretrained checkpoint

Run the repository’s test script with a checkpoint and configuration for a supported benchmark.

Before you start

  • Python 3.7 or later, PyTorch 1.7 or later, CUDA 9.0 or later and GCC 4.9 or later.
  • The dependencies and extensions required by the repository.
  • A pretrained checkpoint and matching configuration, plus the relevant dataset.

Steps

  1. Choose a documented pretrained checkpoint and its corresponding configuration, such as PoinTr on PCN or ShapeNet-55.

  2. Run the command below.

    Documented command
    bash ./scripts/test.sh --ckpts --config --exp_name

  3. For ShapeNet-55, add the documented --mode easy option when evaluating that mode.

Expected outcome

The test script evaluates the selected checkpoint and configuration on the relevant dataset; the README also documents benchmark-specific examples.

Guide limitations

  • The dataset, checkpoint and matching configuration must be available locally.
  • The documented test workflow specifies GPU IDs and requires the project’s runtime environment.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
PointrGenerate completed point clouds from incomplete input samples.LinuxThe repository is public and MIT-licensed. The documented workflows run from the codebase and require a configured Python, PyTorch and CUDA environment.
Pulse Of MotionEstimate a video's physical frame rate from its visual dynamics.LinuxThe repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence.
Neural Image AssessmentTrain the model using AVA images and rating annotations.Cloud, Linux, macOS, Windows
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
yuxumin/PoinTr · Official homepage title · retained in its source language
How we review sourcesSuggest a correction