The feature vectors are passed to the encoder, which learns the image representations using its attention layers.