Datasets:

---

ZehuaZhao

/

SUPERChem

like 5

Thank you for your comment.

Currently, the image column in the Parquet file is stored as STRUCT instead of MAP. This happened because the data type was automatically inferred during export.

In STRUCT mode, Parquet merges all keys (i.e., image paths) that have appeared into a fixed structure, treating each key as an independent field. Since the image paths in different rows are almost all different, this causes the column to logically form a sparse matrix.

For example:

• Row A only has a value in the Path_A field, while the remaining path fields are all null.
• Row B only has a value in the Path_B field, while other path fields are also null.

After verification, all image data is properly stored in the file. The null values currently visible is not an indication of missing data.

thanks for your reply. I can see the images locally, but hope you will fix the issue in the online version.

For a better reading experience, we provide a PDF version of the dataset. Please check it in the Updates of the dataset card.

thank you so much for your open-source contribution. your benchmark is very high quality. I'm currently interning at Alibaba, and I hope we'll have more opportunities for collaboration and exchange in the future.