,

Backbone Can Not be Trained at Once: Rolling Back to Pre-trained Network for Person Re-Identification

,

AutoLR: Layer-wise Pruning and Auto-tuning of Learning Rates in Fine-tuning of Deep Networks