Data&Code

The codes and data for the paper titled “Modeling Multi-Saliency-Based Navigation Cues for Human Navigation in Urban Street Environments".ContentIn the folder Code:-<i>HotspotIdentification.py </i>is used to identify hotspots and regions in the visual saliency map.-<i>StrucutalCalculation.py </i>is used for the calculation of weights.-<i>helper function.ipynb</i> helps reproduce figures and tables in the manuscript.In the folder Data, there are:<i>*.jpg:</i> 30 panoramic images of road intersections.<i>*.xlsx:</i> results obtained from Questionnaire Set 1 and 2.<i>Eye-tracking data</i> is used for fine-tuned the model.*.csv saved data that helps reproduce figures or tables.For POIs, we cannot re-distribute it online due to the T&amp;C, one can request from AutoNavi official website.For 3D city model of study area, one can request from https://portal.csdi.gov.hk/csdi-webpage/, search for <i>3D Spatial Data 3D-BIT00</i>, click preview, and download the data according to the location.How to locate reference inforamtion mentioned in the study:Census Statistics of Hong Kong: open the url, click Major Housing Estates, select Prosperous Garden. Reference data is described in demographic and ecomonic characteristics.Usage:-Visual saliency prediction, given a street scene image:Step 1: perform Image segmentation (Mask2Former) and Saliency Object Prediction (VST) separatelly.Step2: combine above two results through salient object selection strategy for Mask GenerationStep3: perform Image impaitingStep4: execute saliency prediction (MDS-ViTNet), one can fine-tuned the model by using the eye-tracking data we provided.-Semantic saliency prediction, using the POIs data in the study area:Step1: excute POI embeddingStep2: add weighting following *.xlsx provided if your main study groups is the elderly.Step3: identify POIs that belongs to each intersection and excute average pooling.-Structural calculation:Step1: hotspot identification following the <i>HotspotIdentification.py</i>Step2: add weighting follwing the <i>StrucutalCalculation.py</i>