- Newest
- Most votes
- Most comments
It sounds like you're experiencing an issue with your Amazon Rekognition Custom Labels dataset. Let me help troubleshoot this situation.
When working with Amazon Rekognition Custom Labels, there are a few key points to consider:
-
For training an adapter, you need to upload images in bulk for Rekognition to predict labels. The system supports up to 10,000 images, and having more representative images improves results.
-
When you upload images through the console, they are stored in an Amazon S3 bucket - either the default console bucket that Rekognition creates or an external bucket you've specified.
-
If you're using an external S3 bucket, you need specific permissions set up. Make sure your IAM policy includes the necessary S3 permissions (GetBucketAcl, GetBucketLocation, GetObject, GetObjectAcl, GetObjectVersion, GetObjectTagging, ListBucket, PutObject) for your bucket.
-
After uploading images, they need to be labeled. If you're uploading unlabeled images, you'll need to label them through the console before they can be properly counted in your dataset.
-
When using the Rekognition APIs with an SDK (rather than the console), you must provide a manifest file that references images stored in your S3 bucket.
Since your images are showing as uploaded but not being counted, check if:
- The images are in the correct format (PNG or JPEG)
- You've completed the labeling process for the uploaded images
- Your IAM permissions are correctly configured for both Rekognition and the S3 bucket
- If using an external bucket, verify the bucket policy allows Rekognition to access the images
If all these aspects are in order, you might need to try creating a new dataset or contacting AWS support for further assistance.
Sources
Adding more images to a dataset - Rekognition
Step 2: Set up Amazon Rekognition Custom Labels console permissions - Rekognition
Creating training and test datasets with images - Rekognition
Bulk analysis and verification - Amazon Rekognition
answered a year ago
Relevant content
asked 6 years ago
asked 3 years ago
