jun-guan-hct issues

Repositories
Issues
Comments

Results 4 issues of


                                            jun-guan-hct

The parameter amount of the model is larger than that of the original siameserpn?

87M vs 167M

when have two classes, it not work.

layer { name: "loss_cls_pos" type: "FocalLoss" bottom: "rpn_cls_score_reshape" bottom: "labels" propagate_down: 1 propagate_down: 0 top: "loss_cls_pos" include { phase: TRAIN } loss_weight: 1 loss_param { ignore_label: -1 normalize: true }...

how to use Focal-loss when i have two classes

background:0 foreground:1 我用设置 ignore label=0吗?

没看懂您的apply_nms函数

似乎没有排序？但是效果还可以..