Dataset file not correctly labeled for Demo Covid Dataset

Created by Taljinder Kaur, Modified on Fri, 3 Apr at 5:29 PM by Taljinder Kaur

Error message

AL: When looking for the data labels in the parent folder name of covid/nofinding/pneumonia, this file (bkdataset.zip.bkenc) was not correctly labeled

Cause

This error is generated because the dataset was not uploaded to blob storage in the expected folder/label structure or naming convention.

The algorithm expects to infer labels from the parent folder names (e.g., covid, nofinding, pneumonia). If the file bkdataset.zip.bkenc is not placed under the correct directory hierarchy or does not follow the documented structure, label extraction fails.

Impact

Data ingestion and labeling cannot proceed correctly.
Training / inference may fail or produce invalid results due to missing or incorrect labels.

Resolution / Workaround

Review required dataset format
- Refer to the EscrowAI User Manual describing the expected:
  - Folder hierarchy (e.g., /covid, /nofinding, /pneumonia)
  - Naming pattern for encrypted archives (e.g., *.zip.bkenc)
Re-encrypt and re-upload dataset
- Use the Encryption tool within EscrowAI to:
  - Properly encrypt the dataset.
  - Ensure the resulting .bkenc file is placed within the correct parent folder(s) that encode the labels.
- Upload the encrypted dataset to blob storage exactly following the documented specifications.
Verify structure post-upload
- Confirm that:
  - Folder names match the expected label names.
  - Each encrypted archive is in the correct label folder.

Best practices

Before upload, mirror the blob-storage folder structure locally and verify it matches documentation.
Maintain a small “sample dataset” that conforms to the spec and use it as a template.