Open mouxingyang opened 3 years ago
Hi~ I have tried the recommend code using triplet loss and the proposed wrt loss, respectively. However, the one with triplet loss exceeds that with wrt loss by ~7% and ~2.7 in terms of the rank-1 under the POOL and FC setting, respectively.
Hi~ It is a nice job. The proposed WRT loss seems to be the improved version of the original Triplet loss. Did you carry out any ablation studies on the proposed AGW baseline for cross-modality visible-infrared Re-ID? If so, how much performance improved by the WRT comparing to the triplet?
Thanks