最近需要调一个基于Faster R-CNN框架的方法的性能，不可避免的需要调网络中的一些阈值。因此，现在记录一下Faster R-CNN中一些阈值的意义。

1 faster rcnn的一种训练方式

四步训练法

1 stage1_rpn_train
2 stage1_fast_rcnn_train
3 stage2_rpn_train
4 stage2_fast_rcnn_train

2 一些阈值的设置

2.1 stage1_rpn_train.pt

layer {
  name: 'rpn-data'
  type: 'Python'
  bottom: 'rpn_cls_score'
  bottom: 'gt_boxes'
  bottom: 'im_info'
  bottom: 'data'
  top: 'rpn_labels'
  top: 'rpn_bbox_targets'
  top: 'rpn_bbox_inside_weights'
  top: 'rpn_bbox_outside_weights'
  python_param {
     module: 'rpn.anchor_target_layer'
     layer: 'AnchorTargetLayer'
     param_str: "'feat_stride': 16"
   }
}

原理：对产生的anchor进行标记，确定是前景还是背景。通过anchor与ground truth box的overlap来决定是否具有前景。

python层：rpn.anchor_target_layer

labels[max_overlaps >= cfg.TRAIN.RPN_POSITIVE_OVERLAP] = 1
labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0

阈值设置：fast_rcnn.config.py

__C.TRAIN.RPN_POSITIVE_OVERLAP = 0.7
__C.TRAIN.RPN_NEGATIVE_OVERLAP = 0.3

将IOU大于0.7的设置为前景，将IOU小于0.3的设置为背景。

2.2 rpn_test.py

layer {
  name: 'proposal'
  type: 'Python'
  bottom: 'rpn_cls_prob_reshape'
  bottom: 'rpn_bbox_pred'
  bottom: 'im_info'
  top: 'rois'
  top: 'scores'
  python_param {
     module: 'rpn.proposal_layer'
     layer: 'ProposalLayer'
     param_str: "'feat_stride': 8"
  }
}

python层：rpn.proposal_layer

阈值设置：fast_rcnn.config.py

# NMS threshold used on RPN proposals
__C.TRAIN.RPN_NMS_THRESH = 0.7

产生proposals之后的NMS抑制的阈值。

# Number of top scoring boxes to keep before apply NMS to RPN proposals
__C.TRAIN.RPN_PRE_NMS_TOP_N = 12000
# Number of top scoring boxes to keep after applying NMS to RPN proposals
__C.TRAIN.RPN_POST_NMS_TOP_N = 2000

2.3 stage1_fast_rcnn_train.pt

layer {
  name: 'data'
  type: 'Python'
  top: 'data'
  top: 'rois'
  top: 'labels'
  top: 'bbox_targets'
  top: 'bbox_inside_weights'
  top: 'bbox_outside_weights'
  python_param {
     module: 'roi_data_layer.layer'
     layer: 'RoIDataLayer'
     param_str: "'num_classes': 21"
  }
}

原理：对RPN提供的proposals进行处理。选择前景ROI和背景ROI

python层：roi_data_layer.layer

oi_data_layer.minibatch.py
fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH)[0]
bg_inds = np.where((overlaps < cfg.TRAIN.BG_THRESH_HI) &
   (overlaps >= cfg.TRAIN.BG_THRESH_LO))[0]

阈值设置：fast_rcnn.config.py

__C.TRAIN.FG_THRESH = 0.5
__C.TRAIN.BG_THRESH_HI = 0.5
__C.TRAIN.BG_THRESH_LO = 0.1

将IOU大于等于0.5的设置为前景，将IOU大于等于0.1，小于0.5之间的设置为背景。

2.4 其他

阈值设置：fast_rcnn.config.py

# Overlap threshold for a ROI to be considered foreground (if >= FG_THRESH)
__C.TRAIN.FG_THRESH = 0.5

大于0.5的ROI才会被使用bbox 回归

## NMS threshold used on RPN proposals
__C.TEST.RPN_NMS_THRESH = 0.7
## Number of top scoring boxes to keep before apply NMS to RPN proposals
__C.TEST.RPN_PRE_NMS_TOP_N = 6000
## Number of top scoring boxes to keep after applying NMS to RPN proposals
__C.TEST.RPN_POST_NMS_TOP_N = 300

测试阶段，产生proposals的NMS阈值。

# Overlap threshold used for non-maximum suppression (suppress boxes with
# IoU >= this threshold)
__C.TEST.NMS = 0.3

检测完之后对bbox进行NMS抑制的阈值