FileNotFoundError: Can't resolve url, neither hostname or project_dir passed

[CocafeDevBrew]

So, I managed to synchronize my label studio project with a google cloud bucket. I could deploy a ml-backend yolo container with my own pre-trained YOLOv10 model on gcloud run and connect it to label studio as well. But when I try to make predictions, the following error emerges in the logs:

FileNotFoundError: Can't resolve url, neither hostname or project_dir passed: gs://path/to/image/in/bucket

The serviceAccountName in my YAML has all the necessary permissions. What else do I need to do?

[heidi-humansignal]

Hello,

Thank you for contacting Label Studio,

Could you please try to enable **presigned URLs in Label Studio Enterprise? **
In your project’s Settings → Storage, add your Google Cloud Storage bucket as a Source Storage, then enable:

• Treat every object as a source file
• Use pre-signed URLs
  With “Use pre-signed URLs” on, the LS backend will transform gs://… links into HTTPS URLs your ML backend can download. See:
  https://labelstud.io/guide/storage#Google-Cloud-Storage
  Please let us know if it has resolved the issue. Best Regards!

Comment by Oussama Assili
Workflow Run

[CocafeDevBrew]

Hi there,

The "Use pre-signed URLs" was always on. But we don't have Label Studio Enterprise. We work with the self-hosted community edition, straight from GitHub.

[heidi-humansignal]

Hello,

Thanks for the clarification. Can you please try the following suggestions that should potentially address your usecase and let us know!

Sync the bucket locally and use Local Storage integration
If possible, periodically sync your GCS bucket to a folder on your Label Studio server, then use Local Storage integration instead of GCS.
Example sync command:

gsutil rsync -r gs://your-bucket-name /path/to/local/folder

This ensures Label Studio sends plain local file paths to the backend.

• Mount GCS via gcsfuse
  Use gcsfuse to mount your bucket directly into your ML backend container or server. The bucket’s contents will appear as a local folder, so YOLO can read the images without special URL handling.

Comment by Oussama Assili
Workflow Run

[CocafeDevBrew]

Thanks for the suggestion. Also, our LS hosting service only comes with ephemeral storage (apart from the database) and gives us very little control over the underlying file system. I know it would work somehow - i tried it locally - but that defies the entire purpose of having a dedicated cloud storage in the first place. I'll gladly look into gcfuse at some point. Thank you for your time!

[heidi-humansignal]

We are happy to assist you! Quite frankly, we would always recommend using Cloud Storage for seamless integration. However, I wanted to unblock you for the moment while understanding the root cause . usually having Presigned URL ON should be able to allow LS to fetch HTTPs URL instead. Can you please take a screenshot of your Cloud configuration in Label Studio?

Thanks!

Comment by Oussama Assili
Workflow Run

[CocafeDevBrew]

Certainly, here you go:

removed for security

[heidi-humansignal]

Thanks for checking! From the configuration it looks correct.. Could you please try one more workaround to isolate this usecase:

1. Create a new Cloud configuration, ensuring that Presigned URL is enabled.
2. Set up a new project with this configuration.
3. Once done, press Sync and check the tasks again.
   If the issue still persists, please open the Developer Console to see if any errors are being captured. That information will help us share the relevant details with our engineering team.

Thank you!

Comment by Oussama Assili
Workflow Run