论文笔记:Learning Deep Features for Discriminative Localization

旧文赏析 CVPR2016 MIT周博磊
idea很简单,但是很work
对model中判别性区域进行定位(use CAM)
achieve 37.1% top-5 error for object localization on ILSVRC 2014, which is remarkably close to the 34.2% top-5 error achieved by a fully supervised CNN approach.
卷积层能够对物体进行定位,但是FC层丢失了位置信息。replace IT with GAP(global average pooling)


论文笔记:Learning Deep Features for Discriminative Localization_第1张图片

related work:
global average pooling (出自 NIN)


论文笔记:Learning Deep Features for Discriminative Localization_第2张图片

Weakly-supervised object localization:
缺点:not trained end-to-end and require multiple forward passes of a network to localize objects

Class Activation Mapping:

论文笔记:Learning Deep Features for Discriminative Localization_第3张图片

Mc(x,y)直接表示空间位置(x, y)处的activation对于类别c的重要性。

实现起来的话,就是算了加权和。把softmax层和conv5的算了加权 ,得到一个加权的featuremap,然后再把它resize
为什么可以这么做?(公式推导)

以自己的图片为例:label为PDR预测为PDR


论文笔记:Learning Deep Features for Discriminative Localization_第4张图片
论文笔记:Learning Deep Features for Discriminative Localization_第5张图片
论文笔记:Learning Deep Features for Discriminative Localization_第6张图片

可以用在可视化、弱监督学习等等领域
GAP vs GMP
GAP encourages the network to identify the extent of the object
GMP identify just one discriminative part


论文笔记:Learning Deep Features for Discriminative Localization_第7张图片

部分ppt截图从 https://www.youtube.com/watch?v=-Z1NIzLxgRU&t=5s

你可能感兴趣的:(论文笔记:Learning Deep Features for Discriminative Localization)