Dataset · JeffreyXiang
500K 3D assets for generation
JeffreyXiang/TRELLIS-500K
Half a million aesthetically filtered 3D assets for text-to-3D and image-to-3D generation tasks.
What's inside
The dataset contains 500k 3D assets selected from Objaverse(XL), ABO, 3D-FUTURE, HSSD, and Toys4k. Assets were filtered by aesthetic scores; some objects lack text captions. The dataset was used in the article Structured 3D Latents for Scalable and Versatile 3D Generation.
What you can do with it
- Train text-to-3D and image-to-3D generators
- Fine-tune a 3D model on filtered assets
- Evaluate 3D generation quality on a large set
- Tasks
- Image → 3d, Text → 3d
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 2.1K
- Likes
- 72
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("JeffreyXiang/TRELLIS-500K")