File size: 1,784 Bytes
df20198 d063f5f 0d93c3f 01fa056 3e86fbb d063f5f 02e1874 d063f5f b708a9f a444f24 f9ce2b5 19db314 e94936e 8a1c8f5 e94936e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 |
---
tags:
- objaverse
- plyverse
- model
- mesh
- asset
- 3d
- object
pretty_name: Plyverse 1.0
---
# Plyverse 1.0
This is [AllenAI Objverse 1.0](https://huggingface.co/datasets/allenai/objaverse) but stripped of all textures and materials and baked into solid meshes and exported as triangulated PLY files.
The models are all scaled to a unit sphere scale which is a normalised cubic scale then scaled again by 0.55 so that the cubic normal fits within the unit sphere.
The benefits to doing this is reduced file size and memory usage. Can be used to teach neural networks the shapes of objects to then be fed into a re-coloring network.
**This is work-in-progress (WIP)** and I might not even finish it because the process is a little slow, the script I am using to convert the GLB files is a Blender script in Python that seems to leak memory in the second phase for some reason, you can check it out [here](https://huggingface.co/datasets/tfnn/Plyverse-1.0/blob/main/glb_to_ply.py).
Even if it didn't leak... It's still really slow and takes around 48 hours to do a single section, in total there are 160 sections with 5,000 models per section so I will probably just convert the first two sections making a total of 10,000 models.
Refer to the original [AllenAI Objverse 1.0 dataset here](https://huggingface.co/datasets/allenai/objaverse) for the meta-data or you can downloads just the meta-data for PLY-000-000.7z and PLY-000-001.7z [here](https://huggingface.co/datasets/tfnn/Plyverse-1.0/resolve/main/plyverse-meta-data.7z?download=true).
# Sections
- [PLY-000-000](https://huggingface.co/datasets/tfnn/Plyverse-1.0/resolve/main/PLY-000-000.7z?download=true)
- [PLY-000-001](https://huggingface.co/datasets/tfnn/Plyverse-1.0/resolve/main/PLY-000-001.7z?download=true) |