_base_ = '../htc/htc_x101-64x4d_fpn_16xb1-20e_coco.py' # learning policy max_epochs = 28 train_cfg = dict(max_epochs=max_epochs) param_scheduler = [ dict( type='LinearLR', start_factor=0.001, by_epoch=False, begin=0, end=500), dict( type='MultiStepLR', begin=0, end=max_epochs, by_epoch=True, milestones=[24, 27], gamma=0.1) ]