Baseline model (with bottleneck) for person ReID (using softmax and triplet loss).
This is the warmpup strategy learning rate
