Skip to content

CAM的均值化处理 #16

@Fly-Yun-Maker

Description

@Fly-Yun-Maker

您好,感谢您卓越的工作,我仔细阅读了您的文章和代码的一些细节,我有一些不太理解的地方。代码中显示需要对原始的CAM和IS-CAM进行均值化
norm_cam = norm_cam/(F.adaptive_max_pool2d(norm_cam, (1, 1)) + 1e-5) IS_cam = IS_cam / (F.adaptive_max_pool2d(IS_cam, (1, 1)) + 1e-5)
同时,不太理解,为什么需要对特征进行进一步的处理:
`

feature_s = feature_s/(torch.norm(feature_s,dim=1,keepdim=True)+1e-5) # B C (H*W)
`

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions