Where to find updated manifest file after labeling a dataset using a premade manifest for a Rekognition Custom Labels project without training a model?

0

Hi!

I have a training set of images for which I manually created a manifest file respecting the format required to train a Rekognition Custom Labels model for object detection. Both the images and that manifest are stored in an S3 bucket. When creating a project to train the model I noticed that the manifest contained some errors which I corrected manually ("Start labeling" then "Draw bounding boxes") but the updated manifest is nowhere to be found.

I made sure to save the changes and when I browse through the dataset as it appears during the creation of a project I noticed that my changes were taken into account. I want the manifest so that I can store it locally on my computer but I can't seem to find it. Any help is very much appreciated.

Thank you!

1 Answer
0

Unless you specified otherwise, Rekognition Custom Labels will upload everything into the S3 bucket which was created when you first accessed the service. The S3 bucket name will be along the lines of custom-labels-console-<region>-<UID>. Inside this bucket, you should find the following folder structure:

projects/
    your-project-name/
        datasets/
            TRAIN/
                output.manifest
            TEST/
                output.manifest

Take a look at these manifest files. I think these will contain the updates you made.

AWS
S_Moose
answered 10 months ago
  • I thank you for your answer. There is indeed such a folder in the bucket. The issue is I don't have the folder datasets/ (and evidently its child folders TRAIN/ and TEST/ and their files). When I access the your-project-name folder I find a labels.json which is empty.

    I think the structure you're suggesting is only available after training a model.

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions