Object Files, Segmentation, and
Spatiotemporal Continuity
Here are some demonstrations of the various manipulations discussed in the following paper:
Gao, T., & Scholl, B. J. (2010). Are objects required for object-files?: Roles of segmentation and spatiotemporal continuity in computing object persistence. Visual Cognition, 18(1), 82-109.
These demonstrations are provided as Quicktime movies, which can be downloaded or viewed directly in most web-browsers. These movies are a bit large and choppy, but they should be sufficient to illustrate the basic manipulations. If you have any trouble viewing the movies, downloading them and then playing them from your local hard-drive may help. As highly compressed versions of the original stimuli constructed for demonstration purposes, these movies may not preserve the precise spatial and temporal characteristics of the originals.  
Two central tasks of visual processing are (1) to segment undifferentiated retinal images into discrete objects, and (2) to represent those objects as the same persisting individuals over time and motion. In this project we explore the interaction of these two types of processing, in the context of object files -- mid-level visual representations that 'stick' to moving objects on the basis of spatiotemporal properties. Object files can be revealed by object-specific preview benefits (OSPBs), wherein a 'preview' of information on a moving object speeds the recognition of that information at a later point when it appears again on the same object (compared to when it reappears on a different moving object), beyond display-wide priming.  
Experiment #1: Regular Objects (360 KB)  
Many of the experiments reported in this paper employ objects defined solely by their motion -- drawn as random-dot squares on a random-dot background. In order to compare the performance in such conditions with a more typical baseline, however, our first experiment simply measured the robust OSPBs that arise when using objects with explicit contours (visible in each static frame of motion), drawn on a random-dot background. The task in each trial is to press a key to indicate whether the single target symbol presented at the end of the display matches either of the initial two preview symbols.  
Experiments 2 and 5: Motion-Defined Objects (356 KB)  
These conditions involved motion-defined objects, drawn as random-noise patches on a random-noise background. This condition also yielded a robust OSPB -- both when the targets could only appear in the two 'final' locations (in Experiment 3) and when they could also appear in the initial preview locations as well (in Experiment 5). These experiments thus indicate that no static segmentation cues are required to establish object files before the previews appear.  
Experiments 3 and 4: Global Rotation (752 KB)  
These experiments go further than Experiment 2: beyond eliminating static segmentation cues from the initial displays, this experiment removed all cues to distinct objects during the motion: the background rotated along with the objects, rendering them invisible at all times. Thus observers simply perceived a single global random-dot disc that rotated in place, and there were no 'objects', per se. This condition did not yield a reliable OSPB when the targets could appear in a variety of locations, indicating that object files do require segmented objects, and cannot occur in a 'region-based' manner for various locations on larger objects.  
Experiment 6: Testing Postdiction: 500 ms delay (508 KB)  
This experiment was identical to Experiment 5 -- again using motion-defined objects -- but now we introduced a delay between the offset of the previews and the onset of the motion. We obtained a robust OSPB with a 100 ms delay, but not with a 500 ms delay -- despite the fact that these two conditions were otherwise identical, fully randomly interleaved, and tested in the same observers, in the same session. (This movie simulates the 500 ms condition, though the process of exporting the display to a Quicktime animation disrupts the precise timing.) As discussed in the paper, these effects indicate that the preview symbols alone are sufficient to establish the object files in the first place, and that the moving boxes that later appeared in those same locations were then treated as the subsequent stages of those same (preview-) objects. The spatiotemporal continuity between the symbols and the motion-defined boxes sufficed to enable this binding over time (at least with short temporal delays), despite the vast featural differences between the red contours of the preview symbols and the colorless motion-defined squares. This is consistent with the view that surface features may be stored in object files, but only spatiotemporal information directs their maintenance over time.