Models: How could I apply focal loss on Faster RCNN models?

Created on 10 Apr 2018 · 6Comments · Source: tensorflow/models

While I was trying to apply focal loss on Faster RCNN, I got error message like this:

google.protobuf.text_format.ParseError: 28:5 : Message type "object_detection.protos.FasterRcnn" has no field named "loss".

It appears I'm unable to change loss function for Faster RCNN models. In this case, how could I apply focal loss on Faster RCNN models if I really need to? Thank you.

help wanted

Source

AmemiyaYuko

Most helpful comment

@AmemiyaYuko @pkulzc I implemented like this

修改builders/losss_buider.py

     def  build_faster_rcnn_classification_loss(loss_config): 
            ...
            loss_type = loss_config.WhichOneof('classification_loss') 
            # add the code below
            if loss_type == 'weighted_sigmoid_focal':
                 config = loss_config.weighted_sigmoid_focal
                 return losses.SigmoidFocalClassificationLoss(gamma=config.gamma,alpha=config.alpha)
             ...

修改pipeline.config

model{
    faster_rcnn{
         ...
         second_stage_classification_loss{
                 weighted_sigmoid_focal{
                        gamma:2
                        alpha:0.5
                 }
          }
          ...
     }
 }

fryng on 22 Jun 2018

👍6

All 6 comments

Adding code owner for more input. @pkulzc, feel free to add more people to the thread.

qlzh727 on 11 Apr 2018

Currently focal loss is only implemented for SSD. So you can not train Faster RCNN with focal loss by simply changing the config. Contributions are welcomed if you're interested in this implementation.

pkulzc on 10 May 2018

@AmemiyaYuko @pkulzc I implemented like this

修改builders/losss_buider.py

     def  build_faster_rcnn_classification_loss(loss_config): 
            ...
            loss_type = loss_config.WhichOneof('classification_loss') 
            # add the code below
            if loss_type == 'weighted_sigmoid_focal':
                 config = loss_config.weighted_sigmoid_focal
                 return losses.SigmoidFocalClassificationLoss(gamma=config.gamma,alpha=config.alpha)
             ...

修改pipeline.config

model{
    faster_rcnn{
         ...
         second_stage_classification_loss{
                 weighted_sigmoid_focal{
                        gamma:2
                        alpha:0.5
                 }
          }
          ...
     }
 }

fryng on 22 Jun 2018

👍6

@fryng any train or validate report?

bygreencn on 26 Apr 2019

Hi any implementation of faster rcnn+ focal loss?

etatbak on 13 Sep 2019

Hi There,
We are checking to see if you still need help on this, as this seems to be considerably old issue. Please update this issue with the latest information, code snippet to reproduce your issue and error you are seeing.
If we don't hear from you in the next 7 days, this issue will be closed automatically. If you don't need help on this issue any more, please consider closing this.