geodiff 0.9.3

Differentiable geometry representations for shape parameterization and optimization.

Differentiable geometry representations for shape parameterization and optimization.

Project Plan

Stage 1: Initial Setup

• Add Github Actions workflow for Github Pages.
• Create first cut User Docs using Jupyter Books and MyST markdown.
  • What is this package for?
  • Add .gitignore for MyST markdown.
• Launch Github Discussions for the project.
  • Create introductory dicussion post.
• Add MIT License.
• Update pyproject.toml.
  • Maintainers, license, license-file, keywords, classifiers, project urls.
• Add Github Actions workflow for Github Release and PyPI publishing.
• Add CHANGELOG.md to maintain release details.
• Create first tag and push it to initiate first release and publish.

Stage 2: Implement Geometry Representations

• Install necessary dependencies
  • numpy, matplotlib and pytorch.
• Implement loss functions.
  • Start with Chamfer loss.
• Hicks-Henne bump functions.
  • Implement the Hicks-Henne class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• CST parameterization.
  • Implement the CST class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• NICE normalizing flow parameterization.
  • Implement the NICE class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• RealNVP normalizing flow parameterization.
  • Implement the RealNVP class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• NIGnet parameterization.
  • Implement the NIGnet class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• NeuralODE parameterization.
  • Implement the NeuralODE class.
  • Add visualization method.
  • Add type hints and docstrings.
  • Add test script.
  • Add documentation.
  • Merge with main branch.
  • Create a tag and push it to create a release.
• Make Pre-Aux net modular by defining it separately from the invertible networks.
  • Make Pre-Aux net modular for NICE.
  • Change test script for NICE.
  • Make Pre-Aux net modular for all representations.
  • Change test scripts for all representations.
  • Update documentation for all representations.
    • Fix random seed for replicating results.
  • Merge with main branch.
  • Create a tag and push it to create a release.

Stage 3: Implement Latent Vectors for Geometry Representation

• Add latent code functionality to Pre-Aux nets.
• Add latent code functionality to NICE.
• Add test script for training with latent code using autodecoder framework.
  • Use NICE for the test script.
  • Fit two latent codes to fit two differently rotated squares.
• Add latent code functionality to all representations.
• Merge with main branch.
• Create a tag and push it to create a release.

Stage 4: Improve Sampling of Points on the Closed Transform

• Sample points in T [0, 1]^d using Farthest Point Sampling for Blue-noise properties.
  • Write function to compute FPS.
  • Create test script to visualize the point samples with specified number of points.
• Transform T to points on closed manifold to preserve uniform point sampling.
  • Use the arc cosine formulation for transforming t, s to phi, theta in 3D.
  • Update test script to visualize point samples on closed manifold as well.
• Merge with main branch.
• Create a tag and push it to create a release.