Created by: baiyfbupt

With this interface, users can evaluate their models' computational complexity locally before distillation. So as to allocating more reasonable computational resources to different models in subsequent distillation training.

In this case: teacher ResNeXt101_32x16d_wsl model complexity score: 1.940 student ResNet50_vd model complexity score: 0.934 The user should allocate twice(1.940/0.934) the student computing resources for the teacher to get the best performance. It is also consistent with our experimental results.