- This is like an image classification problem, but instead of classifying each image as a class, you classify each pixel as a class
- this is typically done using a decoder/encoder network (like an autoencoder) where the image is shrunk down in dimension, then upscaled back into the same dimension (but the output is the pixel classes)
- techniques to improve performance: