Neurodynamical Saliency WAvelet Model

NSWAM is an open source implementation of a saliency and scanpath models cited below:

Berga, D. & Otazu, X. (2022). A Neurodynamical model of Saliency prediction in V1. (Official) Neural Computation / arXiv preprint -> https://doi.org/10.1162/neco_a_01464

Berga, D. & Otazu, X. (2019). Modeling Bottom-Up and Top-Down Attention with a Neurodynamic Model of V1. (Official) Neurocomputing / bioRxiv preprint -> https://doi.org/10.1016/j.neucom.2020.07.047

RUN

Run "run.m" to run the model to read several images from "input/" folder with "conf/single" parameters.

Otherwise, to run an individual image you must run "saliency_nswam.m" (or "nswam.m" by specifying each parameter), by setting the input RGB file (directly from imread) and its path.

Plot paper figures with "src/plot_metrics_spec_several.m" Plot other paper figures with "src/plot_dynamics.m", "src/plot_dynamics_several.m", "src/plot_scanpaths.m"

Notes (input)

1. Input images are by default in "input/" and in ".png" format.
2. Configuration parameter files are set by default to "conf/single" (NSWAM) or "conf/scanpath" (NSWAM-CM). Configuration parameters are listed in README.txt and described in "src/confgen_all.m"

Notes (output)

1. Output files will be in "output/" with images (by default in ".png") and scanpaths. Other folders such as "gazes", "mean" ... describe the gaze-wise and accumulative gaze-wise saliency maps.
2. Files with dynamical information (membrane time x iteration x channel x scales x orientations x height x width) matrixes will be saved by default in "mats/"

BENCHMARK

Check my saliency benchmark code for running the experiments, or download the saliency maps from my owncloud.

Citing NSWAM and NSWAM-CM:

@article{10.1162/neco_a_01464,
    author = {Berga, David and Otazu, Xavier},
    title = "{A Neurodynamic Model of Saliency Prediction in V1}",
    journal = {Neural Computation},
    volume = {34},
    number = {2},
    pages = {378-414},
    year = {2022},
    month = {01},
    issn = {0899-7667},
    doi = {10.1162/neco_a_01464},
    url = {https://doi.org/10.1162/neco\_a\_01464},
    eprint = {https://direct.mit.edu/neco/article-pdf/34/2/378/1982874/neco\_a\_01464.pdf},
}
@article{BERGA2020270,
    title = {Modeling bottom-up and top-down attention with a neurodynamic model of V1},
    journal = {Neurocomputing},
    volume = {417},
    pages = {270-289},
    year = {2020},
    issn = {0925-2312},
    doi = {https://doi.org/10.1016/j.neucom.2020.07.047},
    url = {https://www.sciencedirect.com/science/article/pii/S0925231220311553},
    author = {David Berga and Xavier Otazu}
}