A big part of using convolutional layers is translation invariance, i.e., features are detected regardless of their position in the image.

In a convolutional encoder-decoder that maps from an image input to, say, an object segmentation, are the features in the deconvolutional/decoding layers still translation invariant? If I were to take the pre-final decoding layer's output, would those features capture the positional information of different objects?

convolutional encoder-decoder architecture


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.