Tensor size issue when running the training script with KITTI

[MaxPRon]

Hello everyone,

I'm trying to the get the script point_pillars_prediction.py. I'm using the kitti dataset and running it as it is. However, after a couple of samples I get the following error message.

tensorflow.python.framework.errors_impl.InvalidArgumentError: Incompatible shapes: [1,252,252,4] vs. [1,150,150,4] [[node gradient_tape/heading_loss/logistic_loss/mul/BroadcastGradientArgs (defined at /code/incenda/3dlidar_object_detection/PointPillar/point_pillars_training_run_kitti.py:69) ]] [Op:__inference_train_function_8839]

Alternatively I have this error message:

tensorflow.python.framework.errors_impl.InvalidArgumentError: logits and labels must be broadcastable: logits_size=[254016,4] labels_size=[90000,4] [[node class_loss/softmax_cross_entropy_with_logits (defined at /code/incenda/3dlidar_object_detection/PointPillar/loss.py:78) ]] [Op:__inference_train_function_8839]

It doesn't depend on one single file but is always after twelve samples. I really can't explain why this is happening.

Previously there was a similar issue related to the tensor shape but I don't think it applies here. Does anyone have some suggestions what the reason for that is or had similiar issues ?

Thanks already for the help :)

Best regards,

Max