Error: Input sets do not have the same labels.

[Continuum200]

I have tried to train a model with over 50,000 images and 227 labels (with bounding boxes) I loaded the images using a csv file with each image left as UNASSIGNED for ML use. 

Each time I try to train it, it gets to the 28 minute mark and errors out with Training pipeline failed with error message: Input sets do not have the same labels.

I am at my wits end trying to work out what is the issue. I cannot find anyone else with the same issue (or I have found one, but there was no solution) and no matter what I do, it errors out at the same point (I have removed some labels etc) 

Does anyone know what the issue is (with more detail) so I can try to fix it? I assume it is talking about the 

Dataset and the Annotation set, but I cannot work out what is causing the issue.

I have followed all bounding box details, but to no avail. Any help would be appreciated. I am on the trial at the moment, so do not want to pay for support until I have 

[ricconoel]

Hi,

What specific product under Vertex AI are you using?Are you training for AutoML Vision or Custom training?

[Continuum200]

AutoML training. I thought it would be a spaces issue(in the label names) , and changed them all to underscores, but that had the same issue. Can a label be too long?

[ricconoel]

This is really difficult to pin point the exact problem if you won't have help with GCP support. Best I could suggest is to make sure you are following the best practices [1] [2] for preparing your training data.

[1] https://cloud.google.com/vertex-ai/docs/image-data/classification/prepare-data

[2] https://cloud.google.com/vertex-ai/docs/image-data/object-detection/prepare-data 

[Continuum200]

I may have to get support.  The best I can think is that of the 270 odd labels,, some have 20 images, some have 2000. That could be what it causing it to fail .

The issue is, I just want support on this one matter, and am happy to pay for it, but have no desire to sign my domain over to Google workspace just to get support on a product that I am trialing to see if it will work for my first project. 

[pocathoughts]

@Continuum200 did you ever manage to resolve this? I'm seeing the same thing also at the 28 minute mark

[davidbr]

Hi, did you solve this problem? I have the same issue and i dont know what else to do.