pixparse/cc3m-wds · Dataset Viewer is failing because splits don't have the same features

The train split has jpg, json and txt while the validation split only has jpg and txt.

The Dataset Viewer expect all the splits to have the same columns.
Should the Viewer automatically add a json column of null values to the validation split in that case ? (merging the features of all the splits)

In the meantime this case be fixed by adding the missing json files to the validation tars, or by specifying the features in the header of the README.md

dataset_info:
  features:
  - name: __key__
    dtype: string
  - name: __url__
    dtype: string
  - name: jpg
    dtype: image
  - name: txt
    dtype: string
  - name: json
    struct:
    - name: caption
      dtype: string
    - name: url
      dtype: string
    - name: key
      dtype: string
    - name: status
      dtype: string
    - name: error_message
      dtype: string
    - name: width
      dtype: int64
    - name: height
      dtype: int64
    - name: exif
      dtype: string
    - name: original_width
      dtype: int64
    - name: original_height
      dtype: int64