Reproduce the results

[lzzzzzm]

Thanks for your impressive work！
I am attempting to reproduce the metrics from the paper. Following the instructions in the README, I evaluated MagicDrive’s performance on downstream tasks. Table 3 in the paper states that MagicDrive achieves 23.14% on Occ Pred, but my results show only 19.4%. Is this reasonable?

[ldkong1205]

@AlanLiangC Please take a look at the above question, thank you