Semantic Segmentation
Label each pixel in the image with a category label
Does not differentiate between different instances
Sliding Window
For every pixel, create a small patch
→ Send patch to CNN to classify pixel
Fully Convolutional Network
Design a network as a bunch of convolutional layers to make predictions for pixels all at once
Expensive!
Design network as a bunch of convolutional layers, with downsampling and upsampling inside the network
Upsampling (Unpooling)
Bed of Nails
Fill with 0s
Nearest Neighbor
Copy nearest neighbor
Bilinear Interpolation
Use two closest neighbors in x and y to construct linear approximations
Bicubic Interpolation
Use three closest neighbors in x and y to construct cubic approximations
- This is how we normally resize images!
Max Unpooling
Pair each max pooling layer with a max unpooling layer
- Remember which position had the max
- Place each max into remembered positions
- Fill rest with 0s
Learnable Upsampling: Transposed Convolution
1D Example
