Abstract
When we view an image for the first time, we rapidly capture its gist, followed by more detailed visual information. Most studies have employed visual recognition or verbal recall tasks to quantify the contents within the gist and details of a visual memory. However, utilizing drawings as a memory measure could reveal more about these representations and their timescales. Here, we conducted three different experiments through Amazon Mechanical Turk to test gist and detail content through drawn memories. The main experiment consisted of a drawing task, where participants (N=120) were exposed to real-world scene images for different lengths of time, varying from 100 ms to 10,000 ms. After a 500 ms delay, they were then asked to draw the image from memory in as much detail as possible. Drawings were monetarily rewarded based on number of objects present in their drawings, to motivate participants. Two separate scoring experiments asked for different participants to determine whether drawings included objects present in the original stimulus (N=410), and whether there were false object insertions not present in the original stimulus (N=91). First, we observed that participants were able to successfully complete memory drawings across all time scales, with no differences in number of successful drawings between the 100 ms and 10,000 ms conditions. Second, false objects were found to be rarely present, with only one false object added to drawings on average in the fastest condition of 100 ms. However, we also observed that more correct objects were drawn from memory in the longest exposure condition, 10 seconds, for all tested images, and retention drastically improved for exposures greater than 500 milliseconds. In sum, while people were generally quite good at conveying the gist of the image at all presentation time scales, the exact accuracy of their stimulus recreation improved with longer exposures.