Comparing different features within the same powerful readout architecture allows us to better understand the relevance of low- versus high-level features in predicting fixation locations, while simultaneously achieving state-of-the-art saliency prediction.
Fixations are colored depending on whether they are better predicted by the high-level deep object features (DeepGaze II) model (blue) or the low-level ...
Understanding Low- and High-Level Contributions to Fixation Prediction
www.computer.org › csdl › iccv
It remains unclear to what extent human fixations can be predicted by low-level (contrast) compared to highlevel (presence of objects) image features.
Comparing different features within the same powerful readout architecture allows to better understand the relevance of low- versus high-level features in ...
The aim of the work is to help establish an improved general method for creating simulations of sufficient fidelity to predict part macro-strengths for various ...
This allows the models to learn nonlinear combinations of the features and fit the scale of the final log density better while still being comparatively.
Comparing different features within the same powerful readout architecture allows us to better un- derstand the relevance of low- versus high-level features in.
People also ask
Do objects predict fixations better than early saliency?
Using DeepGaze and the Mean-Luminance-Contrast model (MLC), we can separate how much low-level and high-level features contribute to fixation selection in ...
[24] M. Kümmerer, T. S. A. Wallis, L. A. Gatys, and M. Bethge, ''Understand- ing low- and high-level contributions to fixation prediction ...
The proposed feedback model can learn abundant contextual representations by bridging a recursive pathway from higher-level feature blocks to low-level layers.