How can I view huggingface dataset images?

[asmith26]

Hi there, I'm trying to view a huggingface dataset including images:

import daft
df = daft.read_parquet("hf://datasets/HuggingFaceM4/DocumentVQA")
df.show(5)

Following https://www.getdaft.io/projects/docs/en/stable/10-min.html#working-with-multimodal-data I try:

df = df.with_column("image_show", daft.col("image").image.decode())

Unfortunately this yields:

DaftCoreException: DaftError::External Unable to create logical plan node.
Due to: DaftError::TypeError ImageDecode can only decode BinaryArrays, got image#Struct[bytes: Binary, path: Utf8]

It looks like the huggingface dataset image column has additional information that needs to handled first, but I can't seem to work out how to fix this?

Many thanks for any help, and this amazing lib! :)

[jaychia]

Oh interesting. Looks like this particular dataset stores images as some kind of struct column, with both the URL and bytes for some reason 🤷

You can try:

df = df.with_column("decoded_image", daft.col("image.bytes").image.decode())

.image.decode() works on a binary column, which in this case is a nested column under the image struct column.

[asmith26]

Many thanks for the very quick response and help @jaychia, can confirm it works!! :)