Zijian Zhou
``` lr: 0.0005 lr_step: '4.83' warmup: true warmup_lr: 0.00005 # typically we will use 4000 warmup step for single GPU on VOC warmup_step: 1000 ``` If I change training dataset,...
  the result of roialign is not correct. but use roipooling is ok.
Thanks for your open code! I tried your method on my own dataset, just set everything as you said step by step. But the result dropped sharply. (70%->22%) The loss...