Supporting data for the submission of Matthew Anderson's thesis: Category and Depth Discrimination in Real-World Scenes (2021)

Contact Matt Anderson (matt.anderson@soton.ac.uk) for further information / if there are any issues. 


----------------------		General Info		----------------------

Thesis doi: https://doi.org/10.5258/SOTON/D1990

ORCID ID (Matt Anderson): 0000-0002-7498-2719

Research funded by a University of Southampton Jubilee Scholarship

Chapters 3-5 are empirical papers. 


----------------------		Chapter 3		----------------------


The data for Chapter 3, Category Systems for Real-World Scenes, is already available here: https://doi.org/10.1167/jov.21.2.8.


----------------------		Chapter 4		----------------------


The data and code for chapter 4 is located in the folder Chapt4_Data. 

The files AggData/SEM_categorization.txt and AggData/STR_categorization.txt contain trial-by-trial response data from the semantic and spatial structure tasks respectively. The columns should be intuitively named, but here is a brief description of each:

- Ppt_No: unique participant identifier
- Scene: scene identifier in the SYNS database (see https://syns.soton.ac.uk/)
- View: view identifier in the SYNS database
- cat_agreement: proportion of participants who selected the ground-truth category, using unlimited viewing durations (see Anderson et al., 2021)
- GT_Category: Ground-truth category. In the semantic file, the numbers 1->6 correspond to Nature, Road, Residence, Farm, Beach, and Car Park / Commercial respectively. In the spatial structure file, numbers 1->4 correspond to Cluttered, Closed Off, Flat, and Tunnel respectively. 
- TrialNumber: order in which the image was presented for a given participant
- imageID: unique identifier for each image
- distance_*: The next 6 variables compute various statistics of the image using coregistered ground-truth LiDAR data. These are the (i) mean across the image, standard deviation, and range. These statistics are also computed for a small patch around the fixation location. 
- distance_bin: median split by the column distance_mean
- Colour_GrayScale: 1 = Grayscale, 2 = Colour
- Stereo_Cond: 1 = Mono, 2 = Stereo, 3 = Stereo-Reversed
- Pres_Time: number of frames, as a multiple of 13.33 msecs (75 Hz refresh rate)
- SceneCat: participant-selected category
- DepthCat: participant-selected depth
- ResponsePeriod: Response time (dunno why I called it period)
- Elapsed: empirical measure of computer flip interval (psychtoolbox)
- Cat_correct: 1 = correct, 0 = incorrect
- Depth_Correct: 1 = correct, 0 = incorrect

The folder AnalysisScripts contains:

mafc_dprime.m
- This script computes dprime for a task with >2 categories. For tasks with 2 categories, it produces d-prime estimates identical to the standard method

RapidSceneCatLMMs_CategoryAnalyses.R
- R script that runs the linear mixed models on the category data. This works on data from both tasks by changing which txt file you load in. 

RapidSceneCatLMMs_DepthAnalyses.R
- Same as above, but for the depth responses


----------------------		Chapter 5		----------------------


The data and code for chapter 5 is located in the folder Chapt5_Data. 

The file AggData/alldata.txt contains trial-by-trial response data from the online and in-lab experiment. The columns should be intuitively named, but here is a brief description of each:

- Ppt: unique participant identifier
- Expt: 1 = online experiment with Prolific participants, 2 = in-lab experiment
- Prestime: Presentation duration in msecs
- gt: ground-truth depth judgement. 0 = left, 1 = right
- Human_resp: human ordinal response. 0 = left, 1 = right
- ID: unique identifier for each image (1800 spanning both experiments)
- ProbeLoc_L_1: Row-location of the first probe in a 800x1200 resized version of the (uncropped) SYNS stereo images
- ProbeLoc_L_2: Column-location of the first probe
- ProbeLoc_R_1: Row-location of the second probe
- ProbeLoc_R_2: Column-location of the second probe
- Scene: scene identifier in the SYNS database (see https://syns.soton.ac.uk/)
- View: view identifier in the SYNS database
- meandepth: mean depth condition
- elev: elevation type (variable or same)
- depthcontrast: binned from 0-200% of the mean depth. Bins have uniform width in log-space
- ddiff: absolute depth difference between the two probes
- depthmu: empirical measurement of the mean depth of the two probe locations
- stereo: stereo condition, 1 = mono, 2 = stereo
- correct: ordinal response correct or not?
- human_respratio: well, this is just the human ratio response
- gtratio: ground-truth depth ratio. 
- RT: response time in msecs


The folder AnalysisScripts contains R scripts that run the linear mixed models on all the above data. The names are self explanatory.