Trellis 3D

Trellis 3D is an open-source framework for building, training, and deploying 3D foundation models with a focus on modularity and reproducible research. It provides a unified interface for working with diverse 3D data types, including point clouds, meshes, and voxel grids, enabling researchers and engineers to experiment with different representations without rewriting core pipelines. The toolkit includes implementations of state-of-the-art 3D architectures, configurable training loops, and utilities for large-scale pretraining and fine-tuning on downstream tasks such as 3D classification, segmentation, object detection, and reconstruction.

A key capability of Trellis 3D is its standardized data handling layer, which streamlines dataset preparation, augmentation, and batching across multiple 3D benchmarks. It integrates evaluation routines, logging, and experiment management to support systematic comparison of models and configurations. The framework is designed to be extensible, allowing users to plug in custom backbones, loss functions, and training strategies. Typical use cases include developing domain-specific 3D models for robotics, autonomous driving, medical imaging, and AR/VR, as well as benchmarking new 3D architectures. By providing consistent abstractions and ready-to-use components, Trellis 3D reduces boilerplate code and helps users focus on model design, analysis, and scientific insights.