A Python implementation of some geometric tools on Kendall 3D shape space for practical applications

This paper introduces a set of Python tools designed to bridge the gap between the theoretical framework of Kendall's 3D shape space and practical computational workflows, addressing specific limitations in existing libraries like Geomstats to facilitate advanced 3D shape analysis.

The Gist

Imagine you are an architect trying to compare the "shape" of different houses. You don't care if a house is big or small (scale), whether it's sitting on the left or right side of the street (position), or if it's rotated to face north or south (orientation). You only care about the geometry: the relative arrangement of the windows, doors, and roof.

This is the core problem the paper tackles. It's about creating a mathematical "map" where every unique house shape has its own address, regardless of how it's moved or resized. This map is called Kendall's 3D Shape Space.

Here is a simple breakdown of what the authors did, using everyday analogies:

1. The Problem: The "Shape Map" is Too Hard to Navigate

The authors explain that while mathematicians have built a perfect, theoretical map for these shapes (using something called Riemannian Geometry), it's like having a map of a mountain range drawn in a language only a few experts speak.

• The Current Tool: There is a popular software library called Geomstats (think of it as a Swiss Army knife for geometry). It's great for general tasks, but if you want to do specific, advanced work with 3D shapes (like analyzing a human skeleton or a building), the Swiss Army knife is missing a few crucial tools.
• The Goal: The authors wrote a new set of "tools" (Python code) to fill those gaps, turning complex math into software that researchers can actually use.

2. The Concept: Flattening the Curved World

To analyze shapes, the authors use a clever trick involving Tangent Spaces.

• The Analogy: Imagine the "Shape Space" is the surface of a giant, curved basketball. If you try to draw a straight line between two points on the ball, it curves. But if you place a flat piece of paper (a tangent plane) touching the ball at just one point, that paper represents a "flat" version of the world right there.
• The Tool: The authors created a function called get_tangent_base_kendall. Think of this as a 3D printer for flat paper. It takes a specific shape (a point on the ball) and generates a flat, grid-like coordinate system (an orthonormal basis) right there. This allows researchers to do simple, flat math (like adding numbers) on the curved shape, then print the result back onto the ball.

3. The Challenge: Measuring the "Curvature"

The paper also tackles Sectional Curvature.

• The Analogy: On a flat sheet of paper, if you draw two parallel lines, they never meet. On a sphere (like the Earth), if you draw two lines starting at the equator going north, they eventually meet at the pole. The "Sectional Curvature" is a number that tells you how much the space is bending in a specific direction.
• The Difficulty: Calculating this for 3D shapes is incredibly hard because the "bending" changes depending on how the shape is rotated. It's like trying to measure how much a piece of clay is twisting while you are spinning it.
• The Solution: The authors developed a specific recipe (in sectional_curvature.py) to calculate this bending.
  1. They break the shape down into its core "skeleton" (using a math trick called SVD, which is like stripping a house down to its frame).
  2. They use a special dictionary of rules (formulas from a book by Kendall) to calculate how the "twisting" parts interact.
  3. They combine these to give you a single number: How curved is the space around this specific shape?

4. Why This Matters (The "So What?")

Why do we need to know if the "shape space" is curvy?

• Medical Imaging: If you are studying how a heart valve changes shape in a patient, you need to know if the changes are "straightforward" or if the mathematical space is curving in a way that distorts your measurements.
• Computer Vision: If a robot needs to recognize a chair from any angle, understanding the curvature of the shape space helps the robot understand that a chair seen from the side is "close" to a chair seen from the front, even if they look very different.

Summary

In short, this paper is a translator's guide.

• The Math: Complex, abstract geometry on curved manifolds.
• The Code: A set of Python tools that act as a bridge.
• The Result: Researchers can now take their 3D shapes, flatten them out to do easy math, measure how "curvy" their world is, and then project the results back, all without needing to be a PhD in pure mathematics.

The authors have essentially taken a high-level, theoretical map and built a GPS and a set of road signs so that practical scientists can actually drive on it.

Technical Summary

Here is a detailed technical summary of the paper "A Python implementation of some geometric tools on Kendall 3D shape space for practical applications."

1. Problem Statement

The analysis of geometric structures in fields like computer vision and medical imaging requires methods that are invariant to position, scale, and orientation. Kendall's Shape Space ($\Sigma^k_3$) provides a rigorous Riemannian manifold framework for this analysis. However, a significant gap exists between the theoretical mathematics of these manifolds and practical computational workflows.

While Geomstats is the leading Python library for Riemannian geometry and manifold-based statistics, it lacks specific utilities required for advanced 3D shape analysis. Researchers currently face hurdles in:

• Generating orthonormal bases for the tangent space of Kendall's 3D shape space to perform simulations.
• Calculating Sectional Curvature, a complex metric specific to the quotient structure of shape space, which is not natively supported in general-purpose libraries.

2. Methodology

The authors propose a set of mathematical derivations and corresponding Python implementations to bridge this gap. The methodology is divided into two primary technical contributions:

A. Orthonormal Basis for Tangent Space Simulation

To simulate data on the shape space, one must work within the Horizontal Tangent Space ($H_Z$) of the pre-shape sphere. The paper details a method to construct an orthonormal basis $\{v_1, \dots, v_d\}$ for this space, where $d = 3k - 7$ (the dimension of the shape space for $k$ landmarks).

• Mathematical Approach: The authors utilize the theory of Riemannian submersions. They decompose the ambient space $R^{3(k-1)}$ into the span of the point $Z$, the vertical distribution (rotations), and the horizontal distribution (the shape space tangent).
• Algorithm:
  1. Define the vertical distribution using the Lie algebra $SO(3)$ basis matrices ($L_x, L_y, L_z$).
  2. Construct a self-adjoint linear operator $f$ whose kernel corresponds to the horizontal space $H_Z$.
  3. Compute the Singular Value Decomposition (SVD) of the matrix representation of $f$.
  4. Select the eigenvectors corresponding to zero eigenvalues to form the orthonormal basis.
• Implementation: This allows researchers to generate random samples in the tangent space and project them back onto the shape manifold.

B. Sectional Curvature Calculation

The sectional curvature $K(u, v)$ measures the deviation of the manifold from Euclidean flatness. For Kendall's 3D shape space, the curvature is non-constant and positive.

• Formula: The curvature is calculated using the formula:
  $$K(u, v) = 1 + \frac{3}{4} \|[u, v]_V\|^2$$
  where $[u, v]_V$ is the vertical component of the Lie bracket of two tangent vectors.
• Algorithm:
  1. SVD Decomposition: Decompose the input shape matrix $X = U\Sigma V^T$ to extract singular values $\lambda_i$.
  2. Basis Construction: Construct a specific adapted basis $\{\frac{\partial}{\partial \lambda_i}, \xi_{ij}\}$ based on the singular values and the matrices $W_{ij}$ derived from Kendall's literature.
  3. Lie Bracket Computation: Apply bilinearity to compute the Lie bracket $[u, v]$. The authors utilize pre-derived commutator formulas (from Kendall et al., 2009) for the basis vectors $\frac{\partial}{\partial \lambda_i}$ and $\xi_{ij}$.
  4. Norm Calculation: Compute the squared norm of the vertical component using inner products defined in the literature.
• Implementation: The code automates the change of basis, the application of complex commutator rules, and the final curvature evaluation.

3. Key Contributions

1. Specialized Python Tools: The authors extend the Geomstats ecosystem by providing specific modules (orthonormal_basis_to_simulate.py and sectional_curvature.py) tailored for 3D shape analysis.
2. Algorithmic Translation: They successfully translate complex theoretical abstractions (Riemannian submersions, Lie algebra brackets on quotient spaces) into efficient, numerical algorithms using SVD and matrix operations.
3. Simulation Capability: The tools enable the generation of random samples in the tangent space, facilitating Monte Carlo simulations and statistical testing on shape manifolds.
4. Curvature Analysis: The implementation provides the first accessible method for computing sectional curvature in Kendall's 3D shape space, a metric previously difficult to compute due to the complexity of the quotient map.

4. Results and Validation

• Simulation Example: The paper demonstrates the utility of the simulate_in_tangent_space function by simulating 3D "houses" defined by 10 landmarks. The code successfully generates variations around a reference shape within the tangent space and projects them back to the manifold.
• Curvature Calculation: The compute_curvature function successfully executes the full pipeline (SVD $\to$ Basis Change $\to$ Bracket Computation $\to$ Curvature) for specific shape configurations, validating the mathematical derivation.
• Software Availability: The source code is made publicly available via a Google Sites repository hosted by Universitat Jaume I, ensuring reproducibility.

5. Significance

This work is significant for the computational geometry and medical imaging communities because:

• Accessibility: It lowers the barrier to entry for researchers wishing to apply advanced Riemannian statistics to 3D shapes without needing to derive complex manifold formulas from scratch.
• Completeness: It addresses a critical missing piece in the Geomstats library, making it a more complete tool for shape analysis.
• Practical Application: By enabling curvature calculation and tangent space simulation, the tools open new avenues for statistical hypothesis testing, shape classification, and understanding the geometric properties of biological or mechanical structures in 3D space.

In summary, the paper serves as a practical bridge between the abstract theory of Kendall's Shape Space and real-world data analysis, providing the necessary software infrastructure to perform rigorous geometric statistics on 3D shapes.