Skip to main content

Table 1 For dataset evaluation, we use the ResNet-50 with triplet loss function and the proposed model with a combination of different loss functions

From: Deep person re-identification in UAV images

Network (dataset)Margin 0.1Margin 0.2Margin 0.3
 mAp (%)Rank-1 (%)mAp (%)Rank-1 (%)mAp (%)Rank-1 (%)
ResNet-50 (ImageNet)53.062.157.865.462.665.0
ResNet-50 (cuhk-sysu)70.872.572.775.570.572.5
ResNet-50 (Market1501)69.770.470.770.669.772.1
ResNet-50 (Cuhk03)71.773.670.571.368.871.7
triplet loss + L-GM loss
ResNet-50 (ImageNet)54.364.259.167.361.464.2
ResNet-50 (cuhk-sysu)68.171.968.273.269.671.1
ResNet-50 (Market1501)67.270.767.172.761.768.7
ResNet-50 (Cuhk03)63.267.067.172.468.871.7
  1. Both networks are trained on one of the existed re-id datasets and fine-tuned on the DRHIT01 dataset. In addition, ResNet-50 is directly fine-tuned from ImageNet on the DRHIT01 dataset. Different margin values are used for triplet loss. The best performing loss at a given margin is presented in italic