Data from "Humans but not Deep Neural Networks Often Miss Giant Targets in Scenes”

The data presented here are from the article "Humans but not Deep Neural Networks Often Miss Giant Targets in Scenes”.

Subject data has been compiled into an accessible format, in either Excel spreadsheets or MATLAB .mat files.

Data from Figure 2 of the paper can be found in "targetDetectionData.mat" and "UnityScenesDNNprobs.xlsx".
In the targetDetectionData.mat file, the data in column 3 ('ContextualCondition') is given a number from 1 to 6, which corresponds to the following conditions: 1 = Normal target size, target present; 2 = Normal target size, target absent; 3 = Control, target present; 4 = Control, target absent; 5 = Mis-scaled target size, target present; 6 = Mis-scaled target size, target absent.

Data from Figure 3 of the paper can be found in "realScenesHumanData.mat" and "realworld_fasterRCNNprobs.mat".

The original Unity images used in the target detection experiment can be found in 'UnityImages.zip'. The additional Unity images that were processed by the deep neural networks are found within 'AdditionalUnityImages.zip' - this folder also contains the subset of the original images that were also processed by the networks (and were used to create the human data plot in Figure 2e).

The responses to the object naming task can be found in "objectNamingData.xlsx", this task used images of isolated targets (from the target detection experiment), which can be found in "objectNamingTaskImages.zip".