topic Re: Displaying images hosted in a Unity Catalog volume in a notebook in Data Engineering

Displaying images hosted in a Unity Catalog volume in a notebook

maasg — Wed, 28 May 2025 10:38:49 GMT

I have an image dataset in a Volume in the Unity Catalog.
As part of exploratory analysis, I want to run some queries and display the resulting image set.

Yet, including dynamic references to images in the Catalog doesn't seem to work on a Databricks Notebook.
This works (and the HTML version as well)
![](/Volumes/catalog/schema/path/img.png)

Yet, if the reference path to the image is dynamic, it doesn't work. For example:
image_path="/Volumes/catalog/schema/path/img.png"
txt_html = f"""img is at:{image_path}"""
displayHTML(txt_html) # Sanity check - OK!
img_html = f"""img src="{image_path}" width="400px"""" # Add the missing braces -- the forum editor doesn't like HTML
displayHTML(img_html) # Doesn't work

How can I dynamically display images from the catalog?

PS: Your post has been changed because invalid HTML was found in the message body. The invalid HTML has been removed. Please review the message and submit the message when you are satisfied.
-=> This deleted my complete post and I had to start over again X3 😞

Re: Displaying images hosted in a Unity Catalog volume in a notebook

Walter_C — Wed, 28 May 2025 12:34:30 GMT

Unity Catalog currently does not fully support displaying image metadata dynamically in notebooks and may trigger errors such as [UC_COMMAND_NOT_SUPPORTED.WITHOUT_RECOMMENDATION]

Suggested workaround:

Consider exporting the images from Unity Catalog to a directory accessible within the cluster (like DBFS) and use dynamic paths to reference images stored directly in DBFS instead. DBFS integration might bypass Unity Catalog limitations for unstructured data.

Re: Displaying images hosted in a Unity Catalog volume in a notebook

maasg — Thu, 29 May 2025 11:59:59 GMT

Hi @Walter_C ,

Thanks for your suggestion. Exporting the images (=> duplicating the storage) is a no-go in our case. We are talking about very large automotive datasets.
Is there another option or should we consider moving this data out of the catalog? We also face limitations from the Delta sharing client perspective to access this data from the ML pipeline.