Map-anything is a 3D scene reconstruction framework and neural geometry estimator designed to transform two-dimensional images into metric three-dimensional spatial representations using feed-forward neural networks. It provides a specialized toolkit for predicting camera intrinsics and ray directions from single images without requiring external geometric metadata.
The project includes a 3D model benchmarking suite that utilizes a unified model wrapper to standardize outputs from diverse reconstruction models. This allows for consistent evaluation and accuracy measurement across various spatial datasets. To facilitate downstream use, it includes a COLMAP data exporter that converts neural reconstruction predictions into formats compatible with photogrammetry and splatting pipelines.
The framework covers a broad capability surface including distributed geometry model training, multi-node cluster orchestration, and inference memory optimization. It also provides tools for metric depth visualization, spatial data standardization, and geometry artifact filtering using normal-based masking.