Kyle1993

Actually I reset the iou_threshold in AnchorTargetCreator and ProposalTargetCreator to make the ratio of positive and negitive be 1:1 almost ``` in AnchorTargetCreator: pos_iou_thresh=0.6, neg_iou_thresh=0.2 in ProposalTargetCreator: pos_ratio=0.8, pos_iou_thresh=0.4, neg_iou_thresh_hi=0.3,...

@FortiLeiZhang I find the same problem and have changed the code, but dosen't work ``` rpn_scores = self.score(h) rpn_scores = rpn_scores.permute(0, 2, 3, 1).contiguous() rpn_scores = F.softmax(rpn_scores.view(n, hh, ww, n_anchor,...

@GaoFangshu Thank you for your reply, your demo is really cool!

@yuanjinsheng sorry,I didn't met the problem, and base on it's source code on github(https://github.com/pytorch/vision/blob/master/torchvision/models/vgg.py), it has the argument 'pretrained' try upgrade your torchvision? btw, here have some bugs in this...

我当时运行倒是没有遇到这个问题，不过google了一下有类似的问题（https://www.gitmemory.com/issue/aditya-vora/FCHD-Fully-Convolutional-Head-Detector/28/482140429） 是不是pytorch更新之后有不兼容？

the bbox is also a positive sample