I ran the inference code with as_two_stage=True in the config file. And there is an error
I see in you code, notably in PSTRHead and PstrTransfomer you provide if else in case we use a as_two_stage Deformable DETR. Although, when I try to run your code with this parameter, it does not run. The PSTRHead.forward method issues an error, "too much values to unpack". Because in the return of self.transfomer you there are two cases and the self.as_two_stage branch misses to return the inter_reid_states variable. With adding the missing variable, the inference is done correctly.
Firstly I create this issue just to notify the small omission.
Secondly, I ask a confirmation that you did not use this Deformable DETR extension - the two stage detector extension - for your paper's result, please? There is an ablation study about one stage versus two stages but, from my understand, it is not about the two stages of the detector.
I ran the inference code with
as_two_stage=Truein the config file. And there is an errorI see in you code, notably in
PSTRHeadandPstrTransfomeryou provideif elsein case we use aas_two_stageDeformable DETR. Although, when I try to run your code with this parameter, it does not run. ThePSTRHead.forwardmethod issues an error, "too much values to unpack". Because in the return ofself.transfomeryou there are two cases and theself.as_two_stagebranch misses to return theinter_reid_statesvariable. With adding the missing variable, the inference is done correctly.Firstly I create this issue just to notify the small omission.
Secondly, I ask a confirmation that you did not use this Deformable DETR extension - the two stage detector extension - for your paper's result, please? There is an ablation study about one stage versus two stages but, from my understand, it is not about the two stages of the detector.