Evaluation Results Differ from README Table

[branda-color]

Hello,

I followed the steps in the README to download the pretrained model ("own model 2") and ran the following command to reproduce the results shown in the figure and table in the README:
python eval.py --path_to_checkpoints=./pretrained_models --path_to_data=./data

Note: I did not include the --grid_calibration_samples=True flag because it significantly increases the evaluation time.

However, the results I obtained differ from the Evaluation table in the README.

My evaluation results:

Test Metric	DataLoader 0
test/offset(k=0)/angular_error	11.520730018615723
test/offset(k=0)/loss	0.024358397349715233
test/offset(k=128)/mean_angular_error	11.36042039937973
test/offset(k=128)/std_angular_error	0.04376071219296659
test/offset(k=9)/mean_angular_error	11.74582172794342
test/offset(k=9)/std_angular_error	0.43741419828264544
test/offset(k=all)/angular_error	11.332329750061035

Expected results from the README:

Has anyone else encountered a similar issue? Is there a solution or any steps I might have missed?

Thank you very much!