def forward()

in easycv/models/loss/set_criterion/set_criterion.py [0:0]

• 56 lines of code
• 17 McCabe index (conditional complexity)

    def forward(self, mask_dict, aux_num):
        """
        compute dn loss in criterion
        Args:
            mask_dict: a dict for dn information
            training: training or inference flag
            aux_num: aux loss number
        """
        losses = {}
        if self.training and 'output_known_lbs_bboxes' in mask_dict:
            known_labels, known_bboxs, output_known_class, output_known_coord, num_tgt = self.prepare_for_loss(
                mask_dict)
            l_dict = self.tgt_loss_labels(output_known_class[-1], known_labels,
                                          num_tgt, 0.25)
            l_dict = {
                k + '_dn':
                v * (self.weight_dict[k] if k in self.weight_dict else 1.0)
                for k, v in l_dict.items()
            }
            losses.update(l_dict)
            l_dict = self.tgt_loss_boxes(output_known_coord[-1], known_bboxs,
                                         num_tgt)
            l_dict = {
                k + '_dn':
                v * (self.weight_dict[k] if k in self.weight_dict else 1.0)
                for k, v in l_dict.items()
            }
            losses.update(l_dict)
        else:
            losses['loss_bbox_dn'] = torch.as_tensor(0.).to('cuda')
            losses['loss_giou_dn'] = torch.as_tensor(0.).to('cuda')
            losses['loss_ce_dn'] = torch.as_tensor(0.).to('cuda')

        if aux_num:
            for i in range(aux_num):
                # dn aux loss
                if self.training and 'output_known_lbs_bboxes' in mask_dict:
                    l_dict = self.tgt_loss_labels(output_known_class[i],
                                                  known_labels, num_tgt, 0.25)
                    l_dict = {
                        k + f'_dn_{i}': v *
                        (self.weight_dict[k] if k in self.weight_dict else 1.0)
                        for k, v in l_dict.items()
                    }
                    losses.update(l_dict)
                    l_dict = self.tgt_loss_boxes(output_known_coord[i],
                                                 known_bboxs, num_tgt)
                    l_dict = {
                        k + f'_dn_{i}': v *
                        (self.weight_dict[k] if k in self.weight_dict else 1.0)
                        for k, v in l_dict.items()
                    }
                    losses.update(l_dict)
                else:
                    l_dict = dict()
                    l_dict['loss_bbox_dn'] = torch.as_tensor(0.).to('cuda')
                    l_dict['loss_giou_dn'] = torch.as_tensor(0.).to('cuda')
                    l_dict['loss_ce_dn'] = torch.as_tensor(0.).to('cuda')
                    l_dict = {
                        k + f'_{i}': v *
                        (self.weight_dict[k] if k in self.weight_dict else 1.0)
                        for k, v in l_dict.items()
                    }
                    losses.update(l_dict)
        return losses