mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml

_BASE_: [
  'mask_rcnn_r50_fpn_1x_coco.yml',
]

pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNeXt101_vd_64x4d_pretrained.pdparams
weights: output/mask_rcnn_x101_vd_64x4d_fpn_2x_coco/model_final

ResNet:
  # for ResNeXt: groups, base_width, base_channels
  depth: 101
  variant: d
  groups: 64
  base_width: 4
  norm_type: bn
  freeze_at: 0
  return_idx: [0,1,2,3]
  num_stages: 4

epoch: 24
LearningRate:
  base_lr: 0.01
  schedulers:
  - !PiecewiseDecay
    gamma: 0.1
    milestones: [16, 22]
  - !LinearWarmup
    start_factor: 0.1
    steps: 1000