Following is smooth L1 Loss layer used in PVANet <https://github.com/sanghoon/pva-faster-rcnn/blob/master/models/pvanet/example_train/train.prototxt> :
layer { name: "rpn_loss_bbox" type: "SmoothL1Loss" bottom: "rpn_bbox_pred" bottom: "rpn_bbox_targets" bottom: "rpn_bbox_inside_weights" bottom: "rpn_bbox_outside_weights" top: "rpn_loss_bbox" include { phase: TRAIN } loss_weight: 1 smooth_l1_loss_param { sigma: 3.0 } } This seems to be a custom loss layer that takes four inputs. I'm trying to use this in MXNet using 'mx.symbol.CaffeLoss'. I see that 'CaffeLoss' only takes 2 inputs (data and label). Does anybody know how I can I use 'CaffeLoss' to use the above loss layer in MXNet? Thanks, Indu