If you’re trying to train object detection, you need to create an object detection project.
Export your dataset with only Auto-Orient added as a preprocessing setting (and no other preprocessing or augmentation applied), be sure the export is in COCO Segmentation format.
From here, unzip the folder, and upload the entire folder to the new object detection project, accept the labels and dataset batch → generate dataset version → export dataset
And here is the training process: ANNOUNCEMENT: Custom Model Weights Upload Overview