http://cvcl.mit.edu/SearchModels GETTING STARTED --------------- %% INSTALLING SearchModels Code %% From the website, this is what you need to do to run the code: (1) Make a new directory on your computer that will contain the all the code you download. (2) Download & unzip these folders into that directory on your computer: "Dataset: Image stimuli" (in Downloads section) "Dataset: EyeData" (in Downloads section) "Target features Maps" (in Downloads section) "Tools" (in Code section) (3) Create a new folder in the same main directory called 'saliencyMaps' (4) Save the file "DemoModels.m" into the same main directory. (5) In Matlab, change your current directory to the main directory and try running "DemoModels". A dialog box will appear and you should be able to run the program without changing any of the defaults. %% RUNNING SearchModels Code %% Run DemoModels.m The dialog box that appears allows you to toggle which maps to display and specify: Which single-model prediction maps to display (Saliency, Target features, Context model) Which combined model prediction maps to display (double-source models, triple-source model) Whether to display the context oracle map Compare two thresholds of map specificity (size of image region) Calculate human agreement? Select: which fixations to DEFINE human agreement region Select: which fixations to TEST human agreement region * REQUIRED folders to download: Tools Target features Maps Dataset: Image stimuli Dataset: EyeData Dataset: Context oracle maps (only if ContextOracle map is selected in Dialog box) * RECOMMENDED folders to download (for speed): Saliency Maps NOTE: If the image does not have a precomputed saliency map, the saliency map is generated and automatically saved in the folder '/saliencyMaps'. To disable this feature, comment line 274 ("DemoModels.m") TROUBLESHOOTING (will be updated to reflect problems as they arise) --------------- 1. Make sure you have downloaded the following folders from the website to a main directory: ("DemoModels.m" and "init.m" should also be in main directory): - Tools - Dataset_STIMULI - Dataset_EyeData - Dataset_ContextOracle (recommended) - saliencyMaps (recommended) - targetfeatureMaps (recommended) - LabelMeToolbox (required if running "TrainContextModel.m") PURPOSE OF ONLINE CODE ---------------------- -1- For an image either (A) selected from the dataset, or (B) novel image* display predictions of: (i) Bottom-up saliency model (Torralba 2003), (ii) Target features model (trained on pedestrian-like features; Dalal & Triggs 2005), (iii) Context model (trained on urban scenes and pedestrian-locations) (iv) Combined predictions: multiple models of seach guidance -2- If image was in original dataset, show the context oracle results (see Ehinger et al) -3- Generate ROC curves for model predictions