Possible bug in mmdet modification?

[maitreygram]

Thanks for your error report and we appreciate it a lot.

Checklist

1. I have searched related issues but cannot get the expected help.
2. I have read the FAQ documentation but cannot get the expected help.
3. The bug has not been fixed in the latest version.

Describe the bug
In mmdet/models/detectors/two_stage.py line 244, there is this snippet of code -

        if self.with_rpn:
            proposal_cfg = self.train_cfg.get('rpn_proposal',
                                              self.test_cfg.rpn)
            for i,x in enumerate(xs):
                rpn_losses, proposal_list = self.rpn_head.forward_train(
                    x,
                    img_metas,
                    gt_bboxes,
                    gt_labels=None,
                    gt_bboxes_ignore=gt_bboxes_ignore,
                    proposal_cfg=proposal_cfg)
                if len(xs) > 1:
                    rpn_losses = upd_loss(rpn_losses, idx=i, weight=loss_weights[i])
                losses.update(rpn_losses)
        else:
            proposal_list = proposals

        for i,x in enumerate(xs):
            roi_losses = self.roi_head.forward_train(x, img_metas, proposal_list,
                                                    gt_bboxes, gt_labels,
                                                    gt_bboxes_ignore, gt_masks,
                                                    **kwargs)
            if len(xs) > 1:
                roi_losses = upd_loss(roi_losses, idx=i, weight=loss_weights[i])                            
            losses.update(roi_losses)

So when self.with_rpn is True and len(xs) > 1, only the proposal list of the last image will be passed to the roi_head repeatedly. Whereas it should be the proposal list of each image respectively.

Bug fix
Please correct me if I am wrong. I think proposal_list should be defined outside the for loop. As the code stands, it can only be used with batch_size=1