Feature request
In datasets library we have options to load image and audio datasets which can be viewed in the Dataset Viewer.
LIKE:
from datasets import load_dataset, Image
from datasets import load_dataset, Audio
https://huggingface.co/docs/datasets/en/image_load & https://huggingface.co/docs/datasets/en/audio_process
I guess there's a need for supporting 3D datasets as well for storing & loading 3D object files like (.glb, .ply, .obj etc...) maybe with something like:
from datasets import load_dataset, Mesh # or `_3D` or whatever is naming convention
Motivation
Multi-Modal datasets are rapidly increasing with modalites like 3D, Video etc, datasets already support experimental Video import.
I have a dataset where I've stored everything as dict so it atleast look balanced across all those 4 columns and 4 modalites It have (Text, Image, Audio, 3D Mesh) but it don't look that good since RAW binary bytes are visible. And Needs to be decoded for that specific field containing the binary data.
Your contribution
Perhaps I could have made a PR for this, but since things might not get much attention here, I shouldn't...
Contributors can try to support it so the the huggingface server is not at heavy duty of loading preview images for each mesh in the dataset viewer that can crash. So a cool svg Icon saying 3D or Mesh inplace of preview can be good. OR they can still try to render an image from those 3D data and use it in preview place just like Image/Video.
Edit:
SO Working on PR for this, gonna push soon...
PUSHED THE PR: #8055
Feature request
In
datasetslibrary we have options to loadimageandaudiodatasets which can be viewed in the Dataset Viewer.LIKE:
I guess there's a need for supporting 3D datasets as well for storing & loading 3D object files like (
.glb, .ply, .objetc...) maybe with something like:from datasets import load_dataset, Mesh # or `_3D` or whatever is naming conventionMotivation
Multi-Modal datasets are rapidly increasing with modalites like 3D, Video etc,
datasetsalready support experimentalVideoimport.I have a dataset where I've stored everything as dict so it atleast look balanced across all those 4 columns and 4 modalites It have (Text, Image, Audio, 3D Mesh) but it don't look that good since RAW binary bytes are visible. And Needs to be decoded for that specific field containing the binary data.
Your contribution
Perhaps I could have made a PR for this, but since things might not get much attention here, I shouldn't...
Contributors can try to support it so the the huggingface server is not at heavy duty of loading preview images for each mesh in the dataset viewer that can crash. So a cool svg Icon saying 3D or Mesh inplace of preview can be good. OR they can still try to render an image from those 3D data and use it in preview place just like Image/Video.
Edit:
SO Working on PR for this, gonna push soon...
PUSHED THE PR: #8055