softmaxloss bottom:score_dif bottom:label_dif top:loss_dif
Relu
1x1-1 conv upsample
pool bottom:conv1_2 top:pool1
...
relu
conv1
conv3_2 bottom:conv3_1top:conv3_2
conv3_3 bottom:conv3_2 top:conv3_3
score_dif bottom:euclidea top:score_dif
conv5_1 bottom:pool4 top:conv5_1
res5c
res3b
fc6 bottom:pool5 top fc6
relu1_1 bottom:conv1_1 top:conv1_1
score bottom:fc7 top:score
pool4 bottom:conv4_3 top:pool4
res2c
fc7 bottom:fc6 top:fc7
conv5_3 bottom:conv5_2 top:conv5_3
pool5 bottom:conv5_3 top:pool5
ReidData top:data top:label
dropout top:fc7
k-grouped 1x1 conv
conv1_2 bottom:conv1_1 top:conv1_2
conv1_1 bottom:data top:conv1_1
shared concatenation
softmaxloss bottom:score bottom: label top:loss
conv3_1 bottom:pool2 top:conv3_1
conv4_3 bottom:conv4_2 top:conv4_3
Droup top:fc6
res4b22
conv4_1 bottom:pool3 top:conv4_1
pool top:pool2
conv2_1 bottom:pool1 top:conv2_1
euclidean bottom:fc7 bottom:label top:euclidean top:label_dif
fc1
conv4_2 bottom:conv4_1 top:conv4_2
fc2
conv2_2 bottom:conv2_1 top:conv2_2
conv5_2 bottom:conv5_1 top:conv5_2
Semantic Supervision
1x1-k conv upsample
pool3 bottom:conv3_3 top:pool3
