SmoothL1Loss

class dragon.vm.caffe.core.layers.SmoothL1Loss(layer_param)[source]

Compute the element-wise error transited from L1 and L2. [Girshick, 2015].

Examples:

layer {
  type: "SmoothL1Loss"
  bottom: "bbox_pred"
  bottom: "bbox_targets"
  bottom: "bbox_inside_weights"
  bottom: "bbox_outside_weights"
  top: "bbox_loss"
  loss_param {
    normalization: BATCH_SIZE
  }
}