Rekognition Custom Labels and Ground Truth integration question

Question

Hello!  
  
I've created a semantic segmentation (SS) job in Ground Truth (GT), and it shows all objects are labeled, and the status is Complete. The manifest file appears to be syntactically correct.   
  
Within Custom Labels (CL), I've created a dataset by choosing "Import images labeled by Ground Truth" option and using the manifest file created in GT. It parses correctly. However, when I look at the available dataset in the console, I see that there are zero labeled images. Viewing the images also displays zero labels.  
  
I have done the same with bounding box (BB) jobs in GT, and they import into CL correctly.   
  
Do CL datasets not support import of SS GT manifest files? I can't find anything in the documentation that states either way.   
  
Thanks for your help.

Answer

Hey lkoivu  
  
Thank you for using CustomLabels!  
  
Unfortunately we do not support semantic segmentation jobs. Currently we support 2 groundtruth formats - Classification Job Output & Bounding Box Job Output.  
  
You can find the formats documented here:  
(a) Sagemaker GT - https://docs.aws.amazon.com/sagemaker/latest/dg/sms-data-output.html  
     - See section "Classification Job Output" for image level labels  
     - See section "Bounding Box Job Output" for bounding box format  
(b) Custom Labels  
     - Image level labels - https://docs.aws.amazon.com/rekognition/latest/customlabels-dg/cd-manifest-files-classification.html  
     - Bounding boxes - https://docs.aws.amazon.com/rekognition/latest/customlabels-dg/cd-manifest-files-object-detection.html  
  
I have passed on feedback to the team to make this more clear in our documentation as well as on our console.  
  
We do have roadmap items to expand our support to other ground truth formats.

Rekognition Custom Labels and Ground Truth integration question

Relevant content