Question/Issue:
I’m having trouble importing a labelled object detection dataset that i am storing in a google cloud storage bucket. When I import it, the images they appear fine but the label files get the error: “Failed to upload /tmp/ei-s3-sync-bmW1sg/training/_annotations.coco.json Missing protected header”.
When importing from a bucket it seems there are only 3 options after the bucket access has been verified: leave unlabelled, label from file name or label from folder name. Object detection needs bounding boxes so the labels are more complex but there doesnt seem to be a way to link to a labels file.
Initially i tried my own dataset, with a COCO JSON label file format. I tried calling the files “bounding_boxes.labels.json”, “bounding_boxes.labels” and “_annotations.coco.json”. I then downloaded the cubes object detection dataset which is used as the example for COCO label format, uploaded that to a google bucket but the labels are still not recognised.
Is there a way for edge impulse to automatically recognise and label object detection datasets when importing them from cloud storage?
Project ID:
719687
Expected Outcome:
EI automatically detected the label file in each of the test and train folders, and applies bounding boxes to images as labels
Actual Outcome:
Files remain unlabelled
Reproducibility:
- [ tick ] Always, but error messages seem to vary for the reason labelled file couldn’t be uploaded
- [ ] Sometimes
- [ ] Rarely
Only tried from Google cloud storage bucket. Using the free tier of edge impulse