Description

The Dataset Viewer crashes when loading datasets containing NIfTI (.nii, .nii.gz) neuroimaging files with the error:

Feature type 'Nifti' not found. Available feature types: ['Value', 'ClassLabel', 'Translation', 'TranslationVariableLanguages', 'LargeList', 'List', 'Array2D', 'Array3D', 'Array4D', 'Array5D', 'Audio', 'Image', 'Video', 'Pdf']

Affected Dataset

https://huggingface.co/datasets/hugging-science/arc-aphasia-bids

This is a 273GB neuroimaging dataset in BIDS format containing NIfTI brain scans.

Root Cause

The datasets library added Nifti feature type support in version 4.4.0 (PR huggingface/datasets#7815), but the Dataset Viewer currently pins datasets==4.1.1 which doesn't include this feature type.

Proposed Solution

1. Bump datasets dependency from 4.1.1 to ^4.4.1
2. Add nibabel ^5.0.0 dependency for NIfTI file handling
3. Implement NIfTI support in the viewer pipeline:
   • Add NiftiSource TypedDict and create_nifti_file() in asset.py
   • Add nifti() handler in features.py
   • Update url_preparator.py for Nifti URL signing
   • Update rows.py to prevent truncation of Nifti columns
   • Update rows_utils.py for multithreaded Nifti uploads

Implementation

I have a working implementation ready with full test coverage. Will submit a PR shortly.

Impact

This would enable viewing of neuroimaging datasets on HuggingFace Hub, supporting the neuroscience research community.