deep learning from scratch chapter 7.cnn
Download as PPTX, PDF1 like43 views
The document discusses convolutional neural networks (CNNs) and their advantages over fully-connected neural networks. It introduces CNN components like convolution and pooling layers. Convolution layers apply filters to input data to extract features while accounting for spatial data structure. Pooling layers downsample data to reduce parameters. The document explains concepts like padding, stride, channels and batches in CNNs. It also discusses implementing CNNs efficiently using im2col to apply filters without for loops for improved performance.
deep learning from scratch chapter 7.cnn
- 1. Feedback
- 2. Interaction Lab. Kumoh National Institute of Technology Deep Learning from Scratch chapter 7. CNN JaeYeop Jeong
- 3. ■Intro ■Convolution Layer ■Pooling Layer ■Implement ■Visualization Agenda Interaction Lab., Kumoh National Institue of Technology 3
- 4. ■Previous neural network Fully-connected : Affine layer • Combines with all neurons in adjacent layers Intro Interaction Lab., Kumoh National Institue of Technology 4
- 5. ■Problem of previous neural network Data is ignored • Image consists of width, height, and channel 3D information • Flatten 3-D data into 1-D data Intro Interaction Lab., Kumoh National Institue of Technology 5
- 6. ■Convolution Neural Network Plus Convolution layer and Pooling Layer • Use 3-D data ■ Advantage of data shape Intro Interaction Lab., Kumoh National Institue of Technology 6
- 7. ■Convolution operation Convolution layer • Filter window • Bias : 1 x 1 • Weight : filter parameter Convolution Layer Interaction Lab., Kumoh National Institue of Technology 7 Input Filter Bias Output
- 8. ■Convolution operation Convolution Layer Interaction Lab., Kumoh National Institue of Technology 8
- 9. ■Padding Fills around the data with specific values Resize the output to large Convolution Layer Interaction Lab., Kumoh National Institue of Technology 9 (4, 4) Padding : 1 (3, 3) Filter (4, 4) Output
- 10. ■Stride Stride : 2 Resize the output to small Convolution Layer Interaction Lab., Kumoh National Institue of Technology 10
- 11. ■ Stride and Padding Input : (H, W), padding : P, stride : S, filter : (FH, FW) • 𝑂𝐻 = 𝐻+2𝑃 −𝐹𝐻 𝑆 + 1 • 𝑂𝑊 = 𝑊+2𝑃 −𝐹𝑊 𝑆 + 1 Input : (4, 4), padding : 1, stride : 1, filter : (3, 3) • 𝑂𝐻 = 4+2 ∗1 −3 1 + 1 = 4 • 𝑂𝑊 = 4+2 ∗1 −3 1 + 1 = 4 Input : (7, 7), padding : 0, stride : 2, filter : (3, 3) • 𝑂𝐻 = 7+2 ∗0 − 3 2 + 1 = 3 • 𝑂𝑊 = 7+2 ∗0 − 3 2 + 1 = 3 Input : (28, 31), padding : 2, stride : 3, filter : (5, 5) • 𝑂𝐻 = 28 + 2 ∗ 2 − 5 5 + 1 = 10 • 𝑂𝑊 = 31 + 2 ∗ 2 −5 5 + 1 = 11 Convolution Layer Interaction Lab., Kumoh National Institue of Technology 11 Value must be integer
- 12. ■3-D data convolution Plus channel • Input channel = filter channel Convolution Layer Interaction Lab., Kumoh National Institue of Technology 12
- 13. ■3-D data convolution Output with multiple channels • Use multiple filters Convolution Layer Interaction Lab., Kumoh National Institue of Technology 13
- 14. ■Batch 3-D data -> 4-D data • (H, W, C) -> (Data, H, W, C) Convolution operation on n data Convolution Layer Interaction Lab., Kumoh National Institue of Technology 14
- 15. ■Pooling layer Not use all data • Pooling window = size of stride ■ 2 x 2 window 2 stride ■ 3 x 3 window 3 stride ■ 4 x 4 window 4 stride Pooling Layer Interaction Lab., Kumoh National Institue of Technology 15 2 x 2 window
- 16. ■Pooling layer No parameter for training Not change channel size Stable Pooling Layer Interaction Lab., Kumoh National Institue of Technology 16
- 17. ■Im2col Not use for loop • Performance : numpy Implement Interaction Lab., Kumoh National Institue of Technology 17 input 4-D data to 2-D data Filter application order
- 18. Implement Interaction Lab., Kumoh National Institue of Technology 18
- 19. Q&A