Heterogeneous pipelines for identification
I am trying to run a face recognition pipline where the pipeline for enrollment and probes are different, basically my enrollment part has one step extra. Currently I have created a new heterogeneous pipeline which takes in two different pipelines as the inputs. I think its not very efficient, since there are common components in both of the pipelines (Iresnet in my case). I think it would be cleaner, if there is a transform which can do things conditionally, meaning, if it sees that the sample is from the enrollment set then apply the transform and skup it if its not from enrollment. Adding the group attribute at the csvloader level might solve this, so that I can do the filtering in this transform.
Any suggestions welcome.