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

Igr

Igr is a Python deep-learning research implementation that learns implicit signed distance functions from raw 2D or 3D point clouds, with or without normals. It includes scripts for training surface reconstructions, generating meshes from checkpoints, evaluating a supplied pretrained shape model, and interpolating latent representations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Igr — 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 IGR when you want a research codebase and shell-driven workflows for fitting signed distance functions to point clouds, producing meshes, or experimenting with D-Faust shape-space learning.

Who it is for

  • Researchers studying implicit neural representations and signed distance functions
  • Computer-vision developers reconstructing surfaces from point-cloud data
  • Practitioners exploring shape-space learning and interpolation with D-Faust scans

Documented uses

  • Train an implicit surface representation from a 2D or 3D point cloud.
  • Evaluate a reconstruction checkpoint to produce a surface mesh.
  • Preprocess D-Faust scans into point clouds with normals.
  • Evaluate the supplied pretrained model on unseen test scans.
  • Interpolate latent representations between shapes.
Before you choose

Requirements & limits

  • The documented environment specifies Python 3.7, PyTorch 1.2, and five additional Python packages.
  • Reconstruction requires editing a setup file and running scripts from a shell.
  • The single-surface workflow accepts 2D or 3D data and supports four listed file formats.
  • The D-Faust workflow includes preprocessing, dataset configuration, training, pretrained evaluation, and latent interpolation as separate stages.

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

Train and mesh a single point-cloud surface

Configure a supported point-cloud input, train an IGR reconstruction, and evaluate a checkpoint to produce a mesh.

Before you start

  • Python 3.7 and PyTorch 1.2
  • NumPy, PyHOCON, Plotly, scikit-image, and trimesh
  • A 2D or 3D point cloud in XYZ, NPY, NPZ, or PLY format

Steps

  1. Set the input path and dimensionality in reconstruction/setup.json, using 2 for 2D data or 3 for 3D.

  2. From the code directory, run python reconstruction/run.py.

  3. Run python reconstruction/run.py --eval --checkpoint CHECKPOINT, substituting an epoch or latest.

Expected outcome

The selected trained checkpoint is evaluated to produce a meshed surface.

Guide limitations

  • The guide covers a single surface reconstruction.
  • The instructions do not specify a supported operating system.
Documentation 1 ↗
Source-based practical guide

Preprocess D-Faust scans for shape-space learning

Convert raw D-Faust scans into point clouds with normals and configure the resulting dataset path for IGR.

Before you start

  • Raw D-Faust scans
  • Python 3.7 and PyTorch 1.2
  • NumPy, PyHOCON, Plotly, scikit-image, and trimesh

Steps

  1. Run python preprocess/dfaust.py --src-path SRC_PATH --out-path OUT_PATH from the code directory.

  2. Set dataset_path in shapespace/dfaust_setup.conf to OUT_PATH/dfaust_processed.

Expected outcome

The scans are preprocessed as point clouds with normals, and the shape-space configuration points to the processed dataset.

Guide limitations

  • This guide covers preprocessing and configuration, not training or evaluation.
  • The documented input dataset is D-Faust.
Documentation 1 ↗
Evidence and transparency

Sources & review

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