BlenderRAG Boosts 3D Object Generation from Text by 30%
BlenderRAG, a retrieval-augmented generation system, improves automatic Blender code synthesis from natural language. Using a curated dataset of 500 expert-validated examples across 50 object categories, it raises compilation success rates from 40.8% to 70.0% and semantic alignment from 0.41 to 0.77 CLIP similarity across four state-of-the-art LLMs. The system requires no fine-tuning or specialized hardware, making it immediately deployable. The dataset and code will be released publicly.
Key facts
- BlenderRAG is a retrieval-augmented generation system for Blender code synthesis.
- It uses a dataset of 500 expert-validated examples across 50 object categories.
- Compilation success rate improved from 40.8% to 70.0%.
- Semantic normalized alignment (CLIP similarity) improved from 0.41 to 0.77.
- Tested across four state-of-the-art LLMs.
- No fine-tuning or specialized hardware required.
- Dataset and code will be available at the provided URL.
Entities
Institutions
- arXiv