edithero benchmark for long-horizon 3d editing

The 3D editing community has traditionally tested methods on a single edit, despite the fact that real-world assets are often built through a sequence of revisions. This limitation is addressed by a new benchmark called EditHero, which focuses on long-horizon, part-level 3D editing. To our knowledge, EditHero is the first benchmark for this purpose. ## Challenges in traditional 3D editing methods Traditional 3D editing methods are typically designed for a single edit, but real-world assets often require multiple revisions to achieve the desired outcome. The complexity of 3D editing tasks makes it challenging to evaluate and improve editing methods reliably. The 3D editing community has traditionally tested methods on a single edit, despite the fact that real-world assets are often built through a sequence of revisions. ## The role of LLM/VLM agents in 3D editing The LLM/VLM agents tend to follow instructions more closely and preserve unedited parts better than non-agentic methods, but each edit can take several minutes. The goal of EditHero is to provide a comprehensive platform for evaluating and improving the reliability of long-horizon, part-level 3D editing. ## The benefits of EditHero The goal of EditHero is to provide a comprehensive platform for evaluating and improving the reliability of long-horizon, part-level 3D editing. EditHero focuses on long-horizon, part-level 3D editing, which is a significant step forward in the development of 3D editing methods.