Inconsistency between Point_rcnn class and RCNNNet one.

[lelouedec]

When the class Point_Rcnn which is if I understand correctly the model class, calls RCNNNet it is using a dict created as folllow :

 rcnn_input_info = {'rpn_xyz': backbone_xyz,
                                   'rpn_features': backbone_features.permute((0, 2, 1)),
                                   'seg_mask': seg_mask,
                                   'roi_boxes3d': rois,
                                   'pts_depth': pts_depth}
if self.training:
  rcnn_input_info['gt_boxes3d'] = input_data['gt_boxes3d']

But in RCNNNet forward the input is used with the following keys and code:

pts_input = input_data['pts_input']
target_dict = {}
target_dict['pts_input'] = input_data['pts_input']
target_dict['roi_boxes3d'] = input_data['roi_boxes3d']
if self.training:
      target_dict['cls_label'] = input_data['cls_label']
      target_dict['reg_valid_mask'] = input_data['reg_valid_mask']
      target_dict['gt_of_rois'] = input_data['gt_boxes3d_ct']

Am I missing something there?

[DaHaiHuha]

Emmm, I met the same pazzel, is there anyone any ideas?

[York1996OutLook]

maybe you could reference /PointRCNN/lib/datasets/kitti_rcnn_dataset.py