[KongchangAI]
Product3DCodeVerse Harness

3DCodeVerse Harness

A framework for 3D content generation and refinement built on a multi-agent directed acyclic graph (DAG). It automatically evaluates and corrects generated programs through execution checks, spatial measurements, and visual feedback, without requiring human feedback at any stage of the refinement loop.

Timeline (last 90 days)

Oct 16

3DCodeVerse is a complete research ecosystem, comprising a large-scale high-quality executable 3D code dataset (3DCodeVerse-2M), a coding framework for spatial reasoning (Harness), and practical guidelines for 3D coding.

Unverified50%
Oct 16

3DCodeVerse Harness is a multi-agent directed acyclic graph (DAG) framework for generating and refining 3D content, automatically evaluating and revising generated programs based on three types of signals: execution checks, spatial measurements and visual feedback.

Unverified50%
Oct 16

The entire generation process of 3DCodeVerse Harness requires no human feedback, and its 'check-correct' loop is completed automatically by agents.

Unverified50%
Oct 16

In paired human evaluation, when driven by Claude Opus 5.5, the 3DCodeVerse harness achieved a 90% win rate against Claude Design (3D Object).

Unverified50%
Oct 16

DAG (directed acyclic graph) in 3DCodeVerse is a way of task orchestration, where each node represents an agent or one step of operation, directed edges represent dependency relationships, and cycles are not allowed because cycles could trap the correction in infinite self-iteration without convergence.

Unverified50%

All Facts (5)

Source Articles