_base_ = './point-rend_r50-caffe_fpn_ms-1x_coco.py' max_epochs = 36 # learning policy 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=[28, 34], gamma=0.1) ] train_cfg = dict(max_epochs=max_epochs)