value error in train.py caused by augment_random_sample_crop

[TheUncivil]

error description
When running the train.py script with a custom dataset, the script crashes immediately with the following message:

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (6,) + inhomogeneous part.

The error did not appear when I set the "augment_random_sample_crop" to False.

Environment:

• OS: Ubuntu 22.04.2
• GPU: NVidia GeForce RTX 2080 Ti
• CUDA Version 11.5.119

[2j1ejyu]

I've got the exast same error. Did you resolved it?

[TheUncivil]

I didn't, I simply set 'augment_random_sample_crop' to false. That means I have less training data augmentation, but that's not currently a problem for me.

[2j1ejyu]

Thanks for replying :) I've solved it by downgrading numpy version

[BBrianZhang]

This issue got resolved when I downgraged numpy to 1.23.0.