The hypothesis that the semantic content of images is critical in guiding eye-movements has recently received support from meaning maps (MMs), a technique developed to be able to capture the distribution of meaning across any given image (see https://www.nature.com/articles/s41562-017-0208-0 …). [2/5]
-
-
Prikaži ovu nit
-
We compare how well MMs and computational saliency models predict fixations and show that DeepGaze II – a deep neural network trained to predict fixations based on high-level features rather than meaning – outperforms MMs (on Y-axes: measures of prediction quality). [3/5]pic.twitter.com/pcIE43K5kB
Prikaži ovu nit -
Next, we show that whereas humans respond to changes in meaning induced by manipulating object-context relationships (we used scenes from SCEGRAM database, created by
@SceneGrammarLab), MMs and DeepGaze II are not able to predict these changes [4/5]pic.twitter.com/iiT445l8Ns
Prikaži ovu nit -
We conclude that meaning maps instead of meaning per se, seem to highlight high-level features that have the potential to carry meaning in typical natural scenes (just like state-of-the-art saliency models). [5/5]
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
congrats Marek, this is awesome!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.