# Hands-on

Initially, we planned to walk through the Marching Cubes algorithm and apply it to the [LGM Demo](https://huggingface.co/spaces/dylanebert/LGM-mini). However, recent advancements in mesh generation have made this approach less relevant.

While a deep dive into the methods behind [MeshAnything](https://huggingface.co/spaces/Yiwen-ntu/MeshAnything) would be much more pertinent, its newness and [non-commercial license](https://github.com/buaacyw/MeshAnything/blob/main/LICENSE.txt) make it suboptimal for the time being.

Instead, here are some resources based on your goals:

- [Splat to Mesh](https://huggingface.co/spaces/dylanebert/splat-to-mesh): If you followed along with with the LGM-based activities and want to produce the final mesh, this open source demo is based on the original [LGM](https://github.com/3DTopia/LGM) codebase. Note that this method is slow and resource-intensive.
- [InstantMesh](https://huggingface.co/spaces/TencentARC/InstantMesh): This is fast and state-of-the-art approach uses FlexiCubes to produce the final mesh. It currently ranks toward the top of the [3D Arena](https://huggingface.co/spaces/dylanebert/3d-arena) leaderboard.
- [meshgpt-pytorch](https://github.com/lucidrains/meshgpt-pytorch): This open source reimplementation of [MeshGPT](https://huggingface.co/papers/2311.15475) provides a good starting point for open-source differentiable mesh generation. [MeshAnything](https://huggingface.co/papers/2311.15475) builds upon MeshGPT. Note: This implementation only provides architecture, not weights.

These resources should help you continue exploring mesh generation and its most recent advancements.

