BioMesh Research

Welcome to the official Hugging Face repository for BioMesh Research. We are an open-source AI research lab dedicated to bridging the gap between evolutionary biology, paleontology, and geometric deep learning.

Our mission is to develop foundational AI models that decode, map, and reconstruct 3D biological geometry across deep time.

Our Focus

We build generative and predictive networks designed to understand the complex mathematical relationships between skeletal structures and full-body morphology.

• Geometric Deep Learning: Processing complex 3D point clouds and triangular meshes directly on hardware.
• Anatomical Reconstruction: Training models to predict soft-tissue volume and outer organic skin envelopes directly from underlying bone coordinates.
• Automated Volumetric Pipelines: Writing custom preprocessing engines to natively transform raw museum CT slice stacks (TIFF/DICOM) into clean, model-ready 3D training assets.

Active Project Ecosystem

1. AlphaMorph (Flagship Model)

An ongoing geometric neural network architecture trained on diverse avian specimens (including high-fidelity museum scans of Alle alle and Gavia immer). The model learns to map variation in rigid bone geometry to full-body surface envelopes.

2. Generalization Benchmarks

Testing model boundaries by feeding structural wildcards—such as heavily compressed amphibian skeletons (Anura) and extinct archosaur lineages—to validate the universal laws of vertebrate geometry.

Technology Stack

• Core Frameworks: PyTorch, PyTorch Geometric
• 3D Data Processing: Open3D, NumPy, OpenCV, Blender Armature Rigging
• Loss Architectures: Symmetric Chamfer Distance, Earth Mover's Distance (EMD)

Collaboration & Data Sourcing

Our training datasets are strictly anchored in anatomically authentic, peer-reviewed scientific data, leveraging massive digital specimen repositories like the openVertebrate (oVert) project and institutional collections.

We welcome collaboration from AI engineers, computational biologists, and paleontologists. If you are interested in synthetic data generation, automated segmentation, or geometric morphology, feel free to explore our repositories or open a discussion space.

Deforming the boundaries of artificial intelligence to reconstruct the history of life.
Models: Coming Soon | Datasets: In Development