The run_models.py file is used to run instances of the elm model with specific configs and to compute any of the three available metrics. Example use:
python run_models.py --d People COCOSearch18 Interiors Unrestricted --m elm_base elm_final nnIBS --mts perf mm hsp
The available configs are elm_base, elm_final and nnIBS, but you can add your own. The available metrics are perf (cumulative performance), mm (multimatch) and hsp (human scanpath prediction). WARNING: hsp takes a long time. The available datasets are those present in the "Datasets" folder
The hyperparamter_search.py file was used to run each experiment in the paper. The results for the experiments run this way are in the "Experiments" folder.
The "Datasets" folder contains all of the datasets used in the paper. Each dataset consists of json files with the human scanpaths and the trials' parameters. The stimuli are not present in this repository, but in each folder there's a link from where they can be downloaded.
The "Model" folder contains the code for the elm model and each of its components.
The "Results" folder contains the files generated by running the "run_models.py" script.
The "Metrics" folder contains all of the code related to the metrics.
The "Heatmaps" folder is used to store plots of the different probability maps such as the posterior.