faster rcnn代码解读（七）rpn_loss

最新推荐文章于 2023-04-17 16:20:05 发布

shchojj

最新推荐文章于 2023-04-17 16:20:05 发布

阅读量356

点赞数

分类专栏： object detection

本文链接：https://blog.csdn.net/fanre/article/details/105385845

版权

object detection 专栏收录该内容

15 篇文章 1 订阅

订阅专栏

faster rcnn代码解读参考

：https://github.com/adityaarun1/pytorch_fast-er_rcnn

https://github.com/jwyang/faster-rcnn.pytorch

其实主要的是bbox的loss、

def smooth_l1_loss(bbox_pred,
                    bbox_targets,
                    bbox_inside_weights,
                    bbox_outside_weights,
                    sigma=1.0,
                    dim=[1]):
    sigma_2 = sigma ** 2
    box_diff = bbox_pred - bbox_targets
    in_box_diff = bbox_inside_weights * box_diff
    abs_in_box_diff = torch.abs(in_box_diff)
    smoothL1_sign = (abs_in_box_diff < 1. / sigma_2).detach().float()
    in_loss_box = torch.pow(in_box_diff, 2) * (sigma_2 / 2.) * smoothL1_sign \
                  + (abs_in_box_diff - (0.5 / sigma_2)) * (1. - smoothL1_sign)
    out_loss_box = bbox_outside_weights * in_loss_box
    loss_box = out_loss_box
    for i in sorted(dim, reverse=True):
        loss_box = loss_box.sum(i)
    loss_box = loss_box.mean()
    return loss_box