Morning Morgan
If I use the sample Jupyter notebook, as is, and train with the public KITTI dataset, I experience the same problems. I then decide to use the tfrecords conversion steps and related spec files, which then works perfectly. It looks like these problems only occurs if I decide to use the “_seg.txt” spec files. We can close this now as I will then be using the tfrecords conversion approach for training.
Thanks for all your assistance