Show HN: Code3DBench – single-image to executable Three.js benchmark

Category: ai-ml

Tags: 3d-generation, benchmark, three-js, multimodal-ai, code-generation

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

Code3DBench is a benchmark for evaluating multimodal models on generating executable Three.js code from a single reference image. It provides a standardized dataset, execution scaffold, and geometry/image metrics, making it interesting for advancing AI-driven 3D content generation and code synthesis evaluation.

Target audience: AI researchers, 3D developers, code generation engineers

Repository: https://github.com/VladimirGl/Code3DBench · Python · MIT

View on Hacker News