Closed 2anchao closed 2 years ago
I carelessness,
books Documentation
* Links to the relevant documentation/comment: Hi code: ``` pos_inds, labels, reg_targets, flattened_hms = \ self._get_ground_truth( grids, shapes_per_level, gt_instances)` logits_pred, reg_pred, agn_hm_pred = self._flatten_outputs( clss_per_level, reg_pred_per_level, agn_hm_pred_per_level)` if self.more_pos: # add more pixels as positive if \ # 1. they are within the center3x3 region of an object # 2. their regression losses are small (<self.more_pos_thresh) pos_inds, labels = self._add_more_pos( reg_pred, gt_instances, shapes_per_level)` ```
questions: why has been use self._get_ground_truth get pos_inds(I call as pos_inds1), and when self.more_pos == True, use self._add_more_pos to get pos_inds(I call as pos_inds2), so the pos_inds2 cover the pos_inds1, I think pos_inds = pos_inds1 + pos_inds2 is right?
I carelessness, the pos_inds1 is None when self.more_pos == True
📚 Documentation
Links to the relevant documentation/comment: Hi code:
pos_inds, labels, reg_targets, flattened_hms = \ self._get_ground_truth( grids, shapes_per_level, gt_instances)`
questions: why has been use self._get_ground_truth get pos_inds(I call as pos_inds1), and when self.more_pos == True, use self._add_more_pos to get pos_inds(I call as pos_inds2), so the pos_inds2 cover the pos_inds1, I think pos_inds = pos_inds1 + pos_inds2 is right?