WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ground truth label problem

elvindp opened this issue · comments

in 'mmdet/models/anchor_heads/solov2_head.py', function 'solov2_target_single' generate ground truth label, but I find that,
when initial cate_label, the value is 0:
cate_label = torch.zeros([num_grid, num_grid], dtype=torch.int64, device=device)
and then update cate_label, if center in that grid:
cate_label[top:(down+1), left:(right+1)] = gt_label
but the class label is starting with 0th class, this will product a lot of 0th class label.
I quite confused. Could anyone help me?