Great work! However, I have a question. After successfully running the code, why is the AUROC_SP metric on MVTec3D about 20 points lower than the result reported in the paper (around 71.9), coming in at less than 50? Additionally, the other two classification metrics, F1_SP and AP_SP, are consistently 2-3 points lower than the results in the paper. Interestingly, when testing on Real3D, the data matches the results reported in the paper, and this data discrepancy does not occur. If you know the reason for this, could you please explain it to me? Thank you!
