Class Activation Map (CAM) Prof. Seungchul Lee Industrial AI Lab. - - PowerPoint PPT Presentation

▶
class activation map cam
SMART_READER_LITE
LIVE PREVIEW

Class Activation Map (CAM) Prof. Seungchul Lee Industrial AI Lab. - - PowerPoint PPT Presentation

Aug 05, 2023 579 likes •702 views

Class Activation Map (CAM) Prof. Seungchul Lee Industrial AI Lab. Issues on CNN (or Deep Learning) Deep learning performs well comparing with any other existing algorithms But works as a black box A classification result is simply

slide-1
SLIDE 1

Class Activation Map (CAM)

  • Prof. Seungchul Lee

Industrial AI Lab.

slide-2
SLIDE 2

Issues on CNN (or Deep Learning)

  • Deep learning performs well comparing with any other existing algorithms
  • But works as a black box

– A classification result is simply returned without knowing how the classification results are derived → little interpretability

  • When we visually identify images, we do not look at the whole image
  • Instead, we intuitively focus on the most important parts of the image
  • When CNN weights are optimized, the more important parts are given higher weights
  • Class activation map (CAM)

– We can determine which parts of the image the model is focusing on, based on the learned weights – Highlighting the importance of the image region to the prediction

2

slide-3
SLIDE 3

Visualizing Convolutional Neural Networks

  • Class Activation Maps (CAMs)
  • A class activation map (CAM) for a given class highlights the image regions used by

the CNN to identify that class

  • B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba.

Learning Deep Features for Discriminative Localization. CVPR'16

3

slide-4
SLIDE 4

Fully Connected Layer

Convolution and pooling layers

Convolutional Neural Networks

Classification Fully connected layer

Convolution Max pooling

7 9

4

slide-5
SLIDE 5

Global Average Pooling

  • Class Activation Map (CAM)
  • (or Attention)

7 9 sigmoid

, x y

Class Activation Map

1

2

k

Convolution and pooling layers

Convolutional Neural Networks

Classification Global Average Pooling

Convolution Max pooling

learned

  • B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba.

Learning Deep Features for Discriminative Localization. CVPR'16

5

slide-6
SLIDE 6

Global Average Pooling Implementation (Naïve)

7 9

sigmoid

1

2

k

7 9 sigmoid

, x y

6

slide-7
SLIDE 7

Global Average Pooling Implementation (Better Way)

1

2

k

7 9 sigmoid

, x y

7

slide-8
SLIDE 8

, x y

Global Average Pooling Implementation (Exactly)

1

2

k

7 9

softmax

8

softmax

, x y

slide-9
SLIDE 9

Example: MNIST

9

slide-10
SLIDE 10

Cantilever

10

In collaboration with KIMM

slide-11
SLIDE 11

Cantilever

In collaboration with KIMM

11