3 ms·
It makes sense to me that attention would be hugely beneficial for vision tasks. We use contextual clues every day to decide what we’re looking at.
by mrfusion 5y ago
It makes sense to me that attention would be hugely beneficial for vision tasks. We use contextual clues every day to decide what we’re looking at.
- sillysaurusx 5y agoIt may make sense, but it also makes no sense. CNNs already have full view of the entire input image. That's how discriminators are able to discriminate in GANs. We added attention and observed no benefits at all in our GAN experiments.