A comprehensive Python system for analyzing, classifying, and retrieving 3D shapes from mesh files using computer vision and machine learning techniques.

GeoVision3D is a complete 3D shape analysis toolkit that enables you to:

• Load and analyze 3D mesh files (OBJ, STL, PLY, GLTF formats)
• Extract geometric features like surface curvature, shape moments, and descriptors
• Calibrate cameras and reconstruct 3D geometry from camera feeds
• Classify and retrieve shapes using machine learning
• Match shapes efficiently using multiple similarity metrics

It combines geometric processing, computer vision, and machine learning in an easy-to-use package.

• Python 3.8 or higher
• pip package manager

1. Navigate to the project directory:

cd GeoVision3D

2. Create and activate virtual environment (recommended):

python -m venv venv
venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

from geovision3d import mesh_utils, geometric_features

mesh = mesh_utils.load_mesh("model.obj")
stats = mesh_utils.get_mesh_stats(mesh)

print(f"Volume: {stats['basic']['volume']:.4f}")
print(f"Surface Area: {stats['basic']['surface_area']:.4f}")
print(f"Sphericity: {stats['descriptors']['sphericity']:.4f}")

import numpy as np
import trimesh
from geovision3d import learned_features, ml_pipeline

# Prepare training data
shapes = [
    trimesh.creation.icosphere(subdivisions=2),
    trimesh.creation.box(),
    trimesh.creation.cylinder()
]
labels = np.array([0, 1, 2])

# Train classifier
extractor = learned_features.LearnedFeatureExtractor()
extractor.fit(shapes)
features = extractor.extract_batch(shapes)

pipeline = ml_pipeline.MLPipeline(normalize=True, reduce_dims=True)
pipeline.train_classifier(features, labels, classifier_type='svm')

# Classify new shape
test_mesh = trimesh.creation.icosphere()
test_features = extractor.extract(test_mesh)
prediction = pipeline.classify(test_features.reshape(1, -1))[0]

from geovision3d import learned_features, matching

# Build database
extractor = learned_features.LearnedFeatureExtractor()
extractor.fit(shapes)
features = extractor.extract_batch(shapes)

retriever = matching.ShapeRetriever(metric='euclidean')
retriever.add_to_database(features, ['sphere', 'cube', 'cylinder'])

# Query
query_features = extractor.extract(test_mesh)
results = retriever.retrieve(query_features, top_k=2)

GeoVision3D/
├── geovision3d/
│   ├── mesh_utils.py              # Load, process, analyze meshes
│   ├── camera_calibration.py       # Camera calibration & distortion
│   ├── reconstruction.py           # 3D reconstruction from images
│   ├── geometric_features.py       # Extract geometric features
│   ├── learned_features.py         # Learn shape descriptors
│   ├── matching.py                 # Shape retrieval & similarity
│   └── ml_pipeline.py              # ML classification & clustering
├── examples/                       # Demo scripts
├── tests/                          # Unit tests
└── requirements.txt

• Mesh processing: Loading, validating, and analyzing 3D mesh files
• Geometric features: Computing curvature, surface normals, shape moments
• Camera calibration: Intrinsic/extrinsic calibration, lens distortion correction
• 3D reconstruction: Stereo depth estimation, point cloud processing, multi-view reconstruction

• Geometric descriptors: Statistical shape analysis, local neighborhood features
• Learned features: PCA-based descriptors, spectral features, histogram-based representations
• Feature learning: Training on shape datasets, dimensionality reduction, normalization

• Classification: Training supervised models (SVM, Random Forest) on shape features
• Clustering: Unsupervised grouping (K-means, hierarchical clustering)
• Retrieval: Efficient shape matching using multiple similarity metrics
• Evaluation: Cross-validation, performance metrics, feature importance

• Object-oriented design: Clean module architecture with single responsibilities
• Type hints & documentation: Type annotations and comprehensive docstrings
• Testing: Unit tests for robust functionality
• Data processing pipelines: Standardized workflows for feature extraction and ML

For camera-based 3D reconstruction:

from geovision3d import camera_calibration, reconstruction

calib = camera_calibration.CameraCalibrator()
K, dist_coeffs = calib.calibrate_from_directory("calibration_images/")

reconstructor = reconstruction.MultiViewReconstructor(K, dist_coeffs)
points_3d, normals = reconstructor.reconstruct_from_sequence("video_frames/")

• Mean Curvature: Discrete mesh curvature estimation
• Gaussian Curvature: Per-vertex curvature analysis
• Shape Descriptors: Moment invariants, PCA-based shape characterization
• Mesh Statistics: Aspect ratio, sphericity, compactness

• Camera Calibration: Zhang's method for checkerboard calibration
• Multi-View Geometry: Structure from Motion (SfM) concepts
• Depth Estimation: Stereo reconstruction and depth refinement
• Point Cloud Processing: Filtering, downsampling, normal estimation

• Feature Normalization: StandardScaler for zero-mean unit-variance
• Dimensionality Reduction: Principal Component Analysis (PCA)
• Classification: Support Vector Machines, Random Forest
• Clustering: K-means, Hierarchical clustering with linkage analysis
• Evaluation: Precision, Recall, F1-score, Silhouette coefficient

• Python 3.8+
• NumPy >= 1.24
• SciPy >= 1.11
• scikit-learn >= 1.3
• OpenCV >= 4.8
• trimesh >= 3.20
• Pillow >= 10.0

• Integration with deep learning (CNN-based feature extraction)
• Real-time streaming 3D reconstruction
• GPU acceleration for large-scale retrieval
• Web API for remote shape analysis
• Interactive 3D visualization module

MIT License

For questions and feedback, please open an issue on the project repository.