Export NDJSON doens't keep original filenames

Support Ultraltyics Platform
1

Hello,

I am using ultralytics platform to edit annotations in a dataset. I can download the annotation via the button on the top-right, which exports a .ndjson file. The file contains images’ names (“file”=…) and their annotations (“annotations”=…), which is exactly what I need. Up untill ~1week ago, the “file” attribute was really the filename, but now it is some sort of serial exadecimal number (? I noticed they start with 00 and proceed increasing), even if the doc continues to say it is the file name.

This is one line from the ndjson file:

{"type":"image","file":"005e19a7098c23e2b8049210e6305723.jpg","url":"https://cdn.ul.run/eu/i/005e19a7098c23e2b8049210e6305723.jpg?Expires=1788971184&KeyName=key-v1&Signature=jD3z-AJ65GevP5nvsCKqdqGMNeE","width":1920,"height":1080,"split":"test","annotations":{"boxes":[[0,0.03452,0.42715,0.01781,0.01707],[0,0.03607,0.59675,0.0054,0.00916]]}}

Is it still possible for the export to either preserve the original filename in file/url (like it used to do) or add a separate field such as original_filename?

(Note that I can still see the original filenames in the web platform while I edit the annotations)

Thanks in advance!

2

Thanks for reporting.

We are investigating this.

3

Hi Daniele, thank you for reporting this and for including such a clear example. We’ve fixed the regression: new Ultralytics Platform NDJSON exports now preserve each image’s original filename and extension in the file field again.

We also added an end-to-end regression test that reads real image metadata and downloads the signed NDJSON export, so CI will fail if filenames are replaced again in the future.

Please let us know if you spot any other issues, or if you have feature ideas that would make the Platform or export workflow better. We really appreciate the feedback!