CreateML Assertion Failure when training Hand Pose model with 5k+ static images

Question

Created Aug ’23

Replies 3

Boosts 4

Participants 4

Hey all, we are currently training a Hand Pose model with the current release of CreateML, and during the feature extraction phase, we get the following error:

Assertion failed: (/AppleInternal/Library/BuildRoots/d9889869-120b-11ee-b796-7a03568b17ac/Library/Caches/com.apple.xbs/Sources/TuriCore/turicreate_oss/src/core/storage/DataTable_data/DataColumn_v2_block_manager.cpp:105): seg->blocks.size()>column_id  [0 > 0]

We have tried to search this online and mitigate the issue, but we are getting nowhere - has anyone else experienced this issue?

Boost

Answer 1

OP

Apple

Aug ’23

Thanks for providing this, we'll look into it. If you could provide the dataset it would be a big help!

0

Answer 2

amlawson98 OP

Dec ’23

I am encountering a similar error. I am using the https://www.kaggle.com/datasets/innominate817/hagrid-classification-512p/ dataset, and subsampling from it.

I also get an error in the pretraining, feature extraction phase: Human Hand Pose Detector was given zero-dimensioned image (0 x 0).

I have tried sampling the first x of each of the 18 classes of handpose images in the datatset. I find that if x = 200 (3600 images) or less, everything runs fine, but if x = 300 (5400 images), I get this error, around the late 4000s of file extraction/prep.

I have checked a couple things:

Every Image in the dataset has been checked to ensure it is not actually a zero dimensional image via a python script with PIL.
I get the same error if I use images 1-200 plus 201-300 as when I use images 1-200 plus 301-400. So i do not believe there is an issue with a particular image.

It seems as though there is a cap on the number of images allowed in a createML workspace, though I am not sure why.

Thank you!

2

Answer 3

bwcooley OP

Jan ’24

Wanted to resurface this thread. I am also experiencing the same issues with Create ML with a 26-class dataset where each class has 3000 images. Similar @amlawson98, Create ML seems to error out in the early 4000s for images. I also followed the same steps for zero-dimensional images. For a dataset with so many classes, the ability to only have 4000 or so images isn't the most ideal as that is a pretty sparse amount of data of each class for the model to train on.

I am using a dataset on Kaggle, but I can't link it here due to forum rules.

Other specs: 16" MacBook Pro with M1 Pro, Create ML Version 5.0 (121.3), Xcode Version 15.1 (15C65).

1