Geo-Algo: Large-Scale Mineralogical Dataset for 3D Reconstruction and Semantic Segmentation (70k Images, 63 Classes)

This dataset contains 70,000 augmented, high-resolution images categorized into 63 distinct mineralogical classes. Developed by the BioPhys-Tech Lab, this collection is engineered to train advanced deep learning architectures (such as ResNet50) for intelligent mineral specimen analysis.

The dataset provides the foundational visual data required for complex geological texture identification, volumetric 3D reconstruction from 2D planes, and precise mineral phase delimitation. It is optimized for workflows involving depth mapping, edge analysis for crystal boundary detection, and spectral clustering for mineral segmentation. This open-access dataset aims to push the boundaries of digital geology and automated material analysis.