In the paper when distilling the new phase, the authors use L2 distillation loss and LD distillation loss.
But in the config file (gfl_r50_fpn_1x_coco_first_40_incre_last_40_cats.py), I only see the LD distillation loss (KnowledgeDistillationKLDivLoss).
so, where is L2 distillation loss?
Thanks in advance.
Thanks for the great work.
In the paper when distilling the new phase, the authors use L2 distillation loss and LD distillation loss. But in the config file (gfl_r50_fpn_1x_coco_first_40_incre_last_40_cats.py), I only see the LD distillation loss (KnowledgeDistillationKLDivLoss).
so, where is L2 distillation loss? Thanks in advance.