About the implement of mist layer

[anlimo1510]

Thanks for your great work.

1. According to your paper, is the process of selecting pseudo boxes in MIST the same as applying NMS with the threshold 0.2 to the top cfg.ROI_BOX_HEAD.OICR_P proposals?

2. In your implement,
   

   wetectron/wetectron/modeling/roi_heads/weak_head/pseudo_label_generator.py

   Lines 35 to 42 in 44e6fa9

   	_boxes = proposals.bbox[max_inds.t().contiguous().view(-1)].view(num_gt_cls.int(), -1, 4)
   	_boxes = BatchBoxList(_boxes, proposals.size, mode=proposals.mode)
   	ious = batch_boxlist_iou(_boxes, _boxes)
   	k_ind = torch.zeros(num_gt_cls.int(), k, dtype=torch.bool, device=device)
   	k_ind[:, 0] = 1 # always take the one with max score
   	for ii in range(1, k):
   	max_iou, _ = torch.max(ious[:,ii:ii+1, :ii], dim=2)
   	k_ind[:, ii] = (max_iou < self.iou_th).byte().squeeze(-1)

   
   max_iou seems return the max iou among the ii proposal and all the proposals with higher score, but not the proposals in the pseudo boxes list. Is it the case?
   And then max_iou will determine whether the ii proposal be appended to the pseudo boxes list. If the max_iou is from the ii proposal and a proposal which is not selected (due to large overlap with existing proposals in pseudo boxes list), then ii proposal will not be appended to the pseudo boxes list. If so, is it slightly different from Algorithm 1 in your paper?

Thank you.