when I download the files from the dataset versions, some IDs do not correspond to the appropriate categories.
The labels have been done manually in the annotations and are correct in the pieces that are coming out wrong.
Let’s say that piece A is pointing to a category that is not the one labeled on that image. And some images are correctly labeled in the _annotations.coco.json file, while others are not.
I tried downloading without augmentations, but the error persists.
This happens to me in versions test06, test05, and test04. I have downloaded previous versions and the same thing is happening.
I also had another problem where a dataset that I had already labeled with id 10073376-004 was deleted, and when I try to upload it, I get a duplicate error. I have deleted the class, but the error persists.
i have other class cuadrado_interno with this id.
Hi @MISAEL_GARCIA!
Happy to help here. I will address your questions in order below:
To ensure I understand, the dataset version you’ve downloaded has annotations corresponding to the incorrect class? Please correct me if I’m wrong.
To help me triage this further, do you grant Roboflow Support permission to access your workspace? If so, can you please provide the name of the project in which this is occurring?
To help me debug this, can you provide a sample image from the set being rejected as duplicates and also the name of the project in which this issue is occurring? Thank you!!
Yes, the labels are different than expected. Today I generated a new version of the dataset called 20251001-test07. This dataset has some images whose labels are incorrect.
Yes, I grant you permission to access my workspace. My project name is ai-part-identifier-intern.
for example
I am looking for an image in the folder/training called DSC04025_JPG.rf.867c2fff68147ce15d52911d6b0e2b19.jp (in roboflow have an label 10061155-002)
this image in train/_annotations.coco.json has these characteristics
image_id: 1664
the image_id 1664 in the annotations has these characteristics
annotation_id: 4387
category_id: 4
category_id 4 is different from the roboflow page
in _annotation.coco.json file have a label 10062464 that is incorrect.
The rejected sample image has the label 10073376. The same project.
I have some problems.
The label is different. But label 10072178-004 and label 278-53-386-1 have the same image.
Label 278-53-81-3 has a different image.
