seanzzzzz/TRELLIS-500K

---
license: mit
task_categories:
- image-to-3d
- text-to-3d
language:
- en
size_categories:
- 100K<n<1M
configs:
- config_name: ObjaverseXL (sketchfab)
data_files:
- split: train
path: "ObjaverseXL_sketchfab.csv"
- config_name: ObjaverseXL (github)
data_files:
- split: train
path: "ObjaverseXL_github.csv"
- config_name: ABO
data_files:
- split: train
path: "ABO.csv"
- config_name: 3D-FUTURE
data_files:
- split: train
path: "3D-FUTURE.csv"
- config_name: HSSD
data_files:
- split: train
path: "HSSD.csv"
- config_name: Toys4k
data_files:
- split: evaluation
path: "Toys4k.csv"
---
# TRELLIS-500K

TRELLIS-500K is a dataset of 500K 3D assets curated from [Objaverse(XL)](https://objaverse.allenai.org/), [ABO](https://amazon-berkeley-objects.s3.amazonaws.com/index.html), [3D-FUTURE](https://tianchi.aliyun.com/specials/promotion/alibaba-3d-future), [HSSD](https://huggingface.co/datasets/hssd/hssd-models), and [Toys4k](https://github.com/rehg-lab/lowshot-shapebias/tree/main/toys4k), filtered based on aesthetic scores.
This dataset serves for 3D generation tasks.
It was introduced in the paper [Structured 3D Latents for Scalable and Versatile 3D Generation](https://huggingface.co/papers/2412.01506).

## Dataset Statistics
The following table summarizes the dataset's filtering and composition:

***NOTE: Some of the 3D assets lack text captions. Please filter out such assets if captions are required.***
| Source | Aesthetic Score Threshold | Filtered Size | With Captions |
|:-:|:-:|:-:|:-:|
| ObjaverseXL (sketchfab) | 5.5 | 168307 | 167638 |
| ObjaverseXL (github) | 5.5 | 311843 | 306790 |
| ABO | 4.5 | 4485 | 4390 |
| 3D-FUTURE | 4.5 | 9472 | 9291 |
| HSSD | 4.5 | 6670 | 6661 |
| All (training set) | - | 500777 | 494770 |
| Toys4k (evaluation set) | 4.5 | 3229 | 3180 |

## Dataset Toolkits

We provide toolkits for data processing in https://github.com/Microsoft/TRELLIS

## More Info

Project page: https://trellis3d.github.io/

Code: https://github.com/Microsoft/TRELLIS