The mAP results implemented in this code are lower than the results reported in the paper #3
Description
I used the code to run on NUS21 following the setting of the paper. But the results on 8/16/24/32 bits are much lower than that reported in the paper. Why is the result so different from that of the paper？
？