Provide Autopilot own data

Question

Is there any way to provide your own (already labeled) data to sagemaker ground truth?  
My plan is to use sagemaker autopilot with my own data that I have already labeled such that the folder structure represents the labels (rekognition provides such functionality).   
The problem is that AutoPilot can only accept manifest files or csv,txt, ect. Well I have images with labels, so my only option is to use a manifest file. HOWEVER GroundTruth create manifest file option only accepts images from only 1 folder without any option to give it labels by default, therefore forcing me to repeat the labeling on an already labeled dataset.  
  
Any idea how to use already labeled data as input???

Answer

Hello...   How are your images labeled?  What format are the labels in?  
  
If you have multiple folders each containing images, and the folder is the label then you could create a quick python (etc) script that will create a manifest file, using the folders as labels.  You can then upload all the folders and the manifest file to S3 and your set.  
  
Groundtruth is a labeling pipeline tool.  Its designed to organize people with domain knowledge to label data.   Also its pretty pricey so you definitely don't want to use it to re-label anything thats already labeled.  
  
Hope that helps.

Provide Autopilot own data

相關內容