architecture: FasterRCNN
use_gpu: true
max_iters: 8000
log_smooth_window: 20
save_dir: output
snapshot_iter: 1000
pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/faster_rcnn_r50_2x.tar
metric: VOC
map_type: 11point
weights: output/faster_rcnn_r50_1x_fruit/model_final
# 3 + 1 (background)
num_classes: 4
finetune_exclude_pretrained_params: ['bbox_pred', 'cls_score']

FasterRCNN:
  backbone: ResNet
  rpn_head: RPNHead
  roi_extractor: RoIAlign
  bbox_head: BBoxHead
  bbox_assigner: BBoxAssigner

ResNet:
  norm_type: affine_channel
  depth: 50
  feature_maps: 4
  freeze_at: 2

ResNetC5:
  depth: 50
  norm_type: affine_channel

RPNHead:
  anchor_generator:
    anchor_sizes: [32, 64, 128, 256, 512]
    aspect_ratios: [0.5, 1.0, 2.0]
    stride: [16.0, 16.0]
    variance: [1.0, 1.0, 1.0, 1.0]
  rpn_target_assign:
    rpn_batch_size_per_im: 256
    rpn_fg_fraction: 0.5
    rpn_negative_overlap: 0.3
    rpn_positive_overlap: 0.7
    rpn_straddle_thresh: 0.0
    use_random: true
  train_proposal:
    min_size: 0.0
    nms_thresh: 0.7
    pre_nms_top_n: 12000
    post_nms_top_n: 2000
  test_proposal:
    min_size: 0.0
    nms_thresh: 0.7
    pre_nms_top_n: 6000
    post_nms_top_n: 1000

RoIAlign:
  resolution: 14
  sampling_ratio: 0
  spatial_scale: 0.0625

BBoxAssigner:
  batch_size_per_im: 512
  bbox_reg_weights: [0.1, 0.1, 0.2, 0.2]
  bg_thresh_hi: 0.5
  bg_thresh_lo: 0.0
  fg_fraction: 0.25
  fg_thresh: 0.5

BBoxHead:
  head: ResNetC5
  nms:
    keep_top_k: 100
    nms_threshold: 0.5
    score_threshold: 0.05

LearningRate:
  base_lr: 0.00001
  schedulers:
  - !PiecewiseDecay
    gamma: 0.1
    milestones: [6400, 7200]
  - !LinearWarmup
    start_factor: 0.
    steps: 100

OptimizerBuilder:
  optimizer:
    momentum: 0.9
    type: Momentum
  regularizer:
    factor: 0.0005
    type: L2

_READER_: 'faster_reader.yml'
TrainReader:
  dataset:
    !VOCDataSet
    dataset_dir: dataset/fruit
    anno_path: train.txt
    with_background: true
    use_default_label: false
  worker_num: -1
  use_process: false

EvalReader:
   inputs_def:
     fields: ['image', 'im_info', 'im_id', 'im_shape', 'gt_bbox', 'gt_class', 'is_difficult']
   dataset:
    !VOCDataSet
    dataset_dir: dataset/fruit
    anno_path: val.txt
    with_background: true
    use_default_label: false

TestReader:
  batch_size: 1
  dataset:
    !ImageFolder
    anno_path: dataset/fruit/label_list.txt
    use_default_label: false
    with_background: true