[KongchangAI]
Benchmark

EditHero

A benchmark for evaluating long-horizon, part-level 3D editing, equipped with natural-language instructions and target images. It employs a deterministic assembly engine combined with human review to systematically assess the iterative 3D model editing capabilities of models.

Timeline (last 90 days)

Oct 5

EditHero is claimed by the research team to be the first benchmark for long-horizon, part-level 3D editing.

Unverified50%
Oct 5

EditHero is equipped with natural language instructions as well as target images that simultaneously cover geometry and texture.

Unverified50%
Oct 5

The core design of EditHero introduces a deterministic assembly engine, enabling the production of precise target results after each edit.

Unverified50%
Oct 5

Every editing sequence in EditHero has been manually reviewed one by one.

Unverified50%
Oct 5

Experiments show that non-agent methods frequently miss changes requested by users while also disturbing regions that should remain unchanged.

Unverified50%
Oct 5

The research team has committed to making the assembly engine and editing sequences publicly available to support reliable iterative 3D editing research.

Unverified50%

All Facts (6)

Source Articles