label-studio-ml-backend
label-studio-ml-backend copied to clipboard
Segment-Anything Unauthorized error when accessing a public resource
I'm encountering a 401 error while utilizing SAM with a remote GCS URL, despite being certain the URL is public. I'm attempting to upload a large JSON file of images via a GCS bucket, but the process fails when invoking the SAM endpoint. It appears the issue stems from this specific line of code: https://github.com/HumanSignal/label-studio-ml-backend/blob/193e24e95539371f53d93d6bee3d8cf684326857/label_studio_ml/utils.py#L66
According to the official documentation, the authentication token should be a Label Studio authentication token. However, using it results in an erroneous Authorization token being added to the request. This is the get_local_path function from the public tools repo: https://github.com/HumanSignal/label-studio-tools/blob/eef952c953fc3c4ed27a276e112ce65f313b7b0a/label_studio_tools/core/utils/io.py#L103
setting access_token
as None
makes remote public calls successful but it fails when calling label-studio
for obvious reasons.
image_local_path = get_local_path(
url=url,
cache_dir=image_cache_dir,
project_dir=project_dir,
hostname=label_studio_host or get_env('HOSTNAME'),
image_dir=image_dir,
access_token=None
)
am I missing any configurations? is this intended?
Thanks :)
@fcuenca4 There is currently a bug preventing the simultaneous use of publicly hosted images and those hosted by Label Studio. This PR will fix the issue.