Dataset · DietCoke4671
Blender Python code generation benchmark for 3D scene editing
DietCoke4671/BlenderBench
A task suite for evaluating AI agents' ability to modify 3D Blender scenes via code based on visual differences.
What's inside
The dataset includes 27 test examples (fewer than 1,000 records) divided into 3 difficulty levels, in English. Each example combines 3D data, source and target scene images, and a text description to test Blender Python script generation.
What you can do with it
- Evaluating multimodal agents on 3D editor code generation tasks
- Testing models on understanding spatial changes between two renders
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 1.3K
- Likes
- 40
How to load it
from datasets import load_dataset
ds = load_dataset("DietCoke4671/BlenderBench")