Seeing Soft Materials Draped Over Objects: A Case Study of Intuitive Physics in Perception,Attention,and Memory

Abstract

We typically think of intuitive physics in terms of high-level cognition, but might aspects of physics also be extracted during lower-level visual processing? Might we not only think about physics, but also see it? We explored this using multiple tasks in online adult samples with objects covered by soft materials—as when you see a chair with a blanket draped over it—where you must account for the physical interactions between cloth, gravity, and object. In multiple change-detection experiments (n = 200), observers from an online testing marketplace were better at detecting image changes involving underlying object structure versus those involving only the superficial folds of cloths—even when the latter were more extreme along several dimensions. And in probe-comparison experiments (n = 100), performance was worse when both probes (vs. only one) appeared on image regions reflective of underlying object structure (equating visual properties). This work collectively shows how vision uses intuitive physics to recover the deeper underlying structure of scenes.

Keywords

intuitive physics perception change detection object-based attention

The human mind is highly attuned to regularities in the world around us (Shepard, 1994), and surely some of the most universal regularities are the laws of physics. As such, psychologists have long been interested in how our minds may (or may not) incorporate physical principles, in the study of “intuitive physics” (for reviews, see Kubricht et al., 2017; Ullman et al., 2017). Classic work, for example, demonstrated that when asked to draw the trajectory of a ball falling out of an airplane in mid-flight, most people fail to draw the correct parabolic paths (McCloskey et al., 1983; see also McCloskey et al., 1980). And such effects generalize to real-world behavior: When asked to drop a ball (while walking) to hit a marked target on the floor, for example, many people release the ball when it is directly over the target, mistakenly predicting that it will fall straight down (McCloskey, 1983).

Perhaps the two most salient themes from this work are (a) that people are often poor at reasoning about such phenomena (with many failing these tasks) and (b) that intuitive physics is centrally a matter of higher-level reasoning and decision-making. (Such errors are thought to stem from “erroneous beliefs” that are held “even after formal training in Newtonian mechanics”; Kaiser et al., 1985, p. 795.)

The current project illustrates how these two themes provide an incomplete picture of how physical regularities are incorporated into the human mind and suggests that although people may often be poor at reasoning about physics, their visual percepts themselves reveal a surprising facility with physical principles (see also Firestone & Scholl, 2017). In short, we may often be poor at thinking about physics, but we may nevertheless also be better at seeing physics. (Of course, the distinction between visual processing and higher-level judgment can be drawn in many ways—e.g., involving automaticity, or stimulus-driven vs. task-dependent processing; for reviews, see Block, 2022; Firestone & Scholl, 2016. We aimed to design experiments that would speak broadly to such differences.) Here, we explored this experimentally in the context of what may at first seem like an unusual domain.

“Cloth Physics”

Intuitive physics is especially salient in phenomena involving colliding billiard balls or collapsing block towers (as reviewed by Kubricht et al., 2017)—but, in fact, physical principles must lie at the root of many other phenomena. An especially fascinating example involves objects being covered by soft materials—as when a chair has a blanket draped over it (e.g., see Tse, 1999, Fig. 23). The visible surfaces of the soft material contain many varied contours with distinctly different physical causes. Some regions (which we will call object regions) will reflect the deep underlying structure of the object itself (i.e., the chair), whereas other regions (which we will call cloth regions) are more superficial (i.e., the blanket’s natural folds and wrinkles, which might differ dramatically each time the same blanket is thrown over the same chair). It is obviously critical for us to apprehend which is which when viewing such scenes, but this can be accomplished only by assessing and appreciating the physical interactions between cloth, gravity, and object (as in the contrast between Figs. 1a and 1b). And indeed, recent computational work confirms that apprehending such relationships requires simulation of physical principles, beyond brute image metrics (Bi et al., 2021). Accordingly, we will refer to such phenomena as “cloth physics.”

Fig. 1.

(a) A cloth-covered object where the top curved section may reflect only the folds of the cloth, caused by gravity. (b) An example in which the top curved section must reflect the structure of the underlying object. (c) The “Veiled Virgin,” an example of a sculpture where observers can readily indicate which contours reflect the object versus soft material (Phillips & Fleming, 2020).

Past work has confirmed that we can distinguish between object regions and cloth regions, as when people must color such regions of stimuli such as Figure 1c differently (Phillips & Fleming, 2020; see also Yildirim et al., 2016, 2022). But such tasks cannot distinguish between perception and thought in the relevant sense: People could succeed either via higher-level reasoning (based on knowledge about cloth, gravity, etc.) or because of how they automatically see such scenes in the first place.

The Current Study

The current project established that cloth physics is taken into account during visual perception itself, using multiple experimental paradigms, each of which had two features that to our knowledge have not been explored in past studies (e.g., Phillips & Fleming, 2020; Ullman et al., 2019; Yildirim et al., 2016, 2022). First, whereas past studies asked for overt judgments about which image contours were which, we employed objective performance-based measures. Second, whereas drawing the cloth/object distinction was the entire explicit goal in past tasks, here this distinction was always entirely task irrelevant. Experiments 1a to 1d explored cloth physics in visual working memory, using change detection, and Experiments 2a and 2b explored cloth physics in visual attention, using probe comparison (inspired by object-based attention studies).

Statement of Relevance

Many times per day, most of us see objects that are covered by soft materials—e.g. a chair with a blanket draped over it. Work in psychological science often starts with experiences like this that we take completely for granted but then shows how such experiences are supported by mental operations that are unexpectedly fascinating or complex. In this work we show that beyond how we may explicitly reason about such familiar layouts, our visual systems themselves infer unexpectedly sophisticated representations from such scenes, spontaneously extracting and highlighting the structure of the covered objects by taking into account subtle physical interactions between cloth, gravity, and object. This shows how the seemingly simple act of perceiving (and attending, and remembering) cloth-covered objects involves a surprisingly elaborate analysis of “intuitive physics.”

This study was performed in line with the principles of the Declaration of Helsinki. All experimental methods and procedures were approved by the Yale University Institutional Review Board. Informed consent was obtained from all individual participants included in the study.

Experiment 1a: Sequential Change Detection

Observers saw two images of cloth-covered objects appear quickly, one after the other, and simply had to detect whether the two raw images were identical. As depicted in Figure 2a, image changes could involve either (a) a new draping of the cloth (as if the cloth were thrown over the same object again, with substantive changes to cloth regions) or (b) changing the object under the cloth (with substantive changes to object regions). Critically, the sheer amount of visual change was always greater in the first condition than the second—in terms of both the brute number of pixels changed and also the degree of higher-level feature change (as quantified from relatively late layers in a convolutional neural network trained for object recognition; VGG16; Simonyan & Zisserman, 2015). We expected better detection for changes to object regions versus cloth regions, despite the greater degree of visual change in the latter.

Fig. 2.

(a) Depictions of the two key conditions in Experiments 1a and 1b: Images could undergo changes either to the underlying objects or to the superficial folds of the cloths. (b) Average accuracy in each condition of Experiment 1a. (c) Average accuracy in each condition of Experiment 1b. (d) Depictions of the matched key conditions from Experiments 1c and 1d. (e) Average accuracy in each condition of Experiment 1c. (f) Average accuracy in each condition of Experiment 1d. In all graphs, error bars depict 95% confidence intervals. Asterisks indicate significant differences between conditions (*p < .05, ***p < .001). ISI = interstimulus interval.

Method

Participants

Two hundred observers (79 female; mean age = 24.55 years) participated for monetary compensation using the Prolific online platform (Palan & Schitter, 2018), and this preregistered sample size was determined before data collection began. Observers were excluded (with replacement) according to preregistered criteria if they reported (in response to postexperimental debriefing questions) that the total number of images they saw appear on screen throughout the entire session was anything other than two images (n = 8) or if they took longer than 10 s to respond (n = 22). All reported experiments employed protocols that were reviewed and approved by the Yale University Institutional Review Board.

Apparatus

After agreeing to participate, observers were redirected to a website where stimulus presentation and data collection were controlled via custom software written using a combination of HTML, Cascading Style Sheets (CSS), JavaScript, Hypertext Preprocessor, and the JsPsych libraries (de Leeuw, 2015). Observers completed the experiment on either a laptop or desktop computer. (Because the experiment was rendered on observers’ own web browsers, viewing distance, screen size, and display resolutions could vary dramatically, so we report stimulus dimensions below using pixel values.)

Stimuli

All text, across the instructions and prompts, was presented in a modified version of jsPsych’s default CSS style: light gray (No. D3D3D3) text (on a black background) drawn in the Open Sans font, presented at a font size of 18 pixels.

A single Tetris-like object was first created in Blender (Version 2.83; https://www.blender.org/), with a 1.2 m × 0.2 m × 0.2 m rectangular prism for a trunk and a 0.4 m × 0.2 m × 0.2 m rectangular prism for a branch. These two prisms were then used to construct six base-objects by first positioning the branch at a 0.05-m offset from the center of the trunk and then randomly rotating the resulting object (six separate times) along its x, y, and z axes. Six corresponding modified-objects were then created by holding these rotations constant but shifting the branches to have a 0.35-m offset (rather than only a 0.05-m offset) from the center of the trunk.

Objects were depicted with white cloths draped over them (on a black background); this draping was simulated by a particle-based physics engine (Nvidia FleX; Macklin et al., 2014). The base-object was centered at the origin, with a piece of cloth mesh dropped from a height (h) of 1.0 m from one of 10 possible angles (0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4, 1.6, 1.8, and 2.0 rad). The cloth was simulated as a grid of particles connected to each other by massless springs that collectively simulated stiffness; the cloth itself was 64 × 64 particles (each particle having a radius of 0.01 m), and cloth mass and all stiffness values (bending, shearing, stretching) were held at a constant of 1.0. The initial velocity and position of the cloth mesh differed according to each angle (α): The initial velocity of the cloth (v_x, v_y, v_z) was set to be (1.8cos(α), 1.8sin(α), 0), and the initial position of the cloth (p_x, p_y, p_z) was (c_x-0.8v_xsqrt(2h × 9.8), c_y-0.8v_ysqrt(2h × 9.8), 0), where c _x and c_y refer to the x and y positions, respectively, of the base-object’s center of mass. During the simulation, the cloth mesh unfolded over 200 consecutive time steps under the influence of gravity, and each time step (frame) was 1/60 s. The output of the simulation from the last time step (i.e., 200th frame) was rendered in grayscale via Blender (on a black 400- × 400-pixel background), and the visible cloth shapes (as in the examples from Fig. 2a) were presented with an average area of 132.89 × 241 pixels, with widths ranging from 104 to 172 pixels and heights ranging from 209 to 267 pixels—these values varied on the basis of the specific orientation of each base-object.

Six corresponding redraped-objects were then created from the base-objects by holding the rotations and branch positions constant but redraping the cloth from a different angle—which in practice produced different superficial cloth shapes (as in the contrast between the base image and the cloth-change panels of Fig. 2a).

Procedure and design

Observers each viewed a single trial during which two images were presented, one after the other, for 850 ms each (separated by a 750-ms blank interval). Images appeared 170 pixels away from the center of the screen (as measured from the center of the image), and each image could appear in one of four locations (and never with both images appearing in the same location)—either to the upper right (45°), the lower right (135°), the lower left (225°), or the upper left (315°) of the screen’s center. Their task (as instructed before the images were presented) was simply to indicate via a key press whether the two images were identical. Both images consisted of cloth-covered objects as described above. On cloth-change trials, the second image was of a redraped-object. On underlying-object-change trials, the second image was of a cloth-covered modified-object. Examples of both trial types are depicted in Figure 2a.

Critically, the magnitude of visual change on a cloth-change trial was always greater than on its corresponding underlying-object-change trial (where these two trials were viewed by different observers). This was always true in terms of both (a) the number of changed pixels in the images themselves and (b) the change in higher-level visual features, quantified by calculating squared Euclidean distance in vectorized feature-activation maps from the second-to-last layer (layer fc1) in a convolutional neural network pretrained for image classification and detection (VGG16; Simonyan & Zisserman, 2015)—with average distances of 7,459.15 for cloth-change trials and 5,630.75 for underlying-object-change trials. (And Fourier analyses also revealed no systematic spatial frequency differences across the images in cloth-change trials vs. their corresponding images in underlying-object-change trials.)

Results

As depicted in Figure 2b, changes were detected much more accurately on underlying-object-change trials (62.00%) compared with cloth-change trials (44.00%), χ² (N = 200) = 6.50, p = .011, effect-size index w = 0.18.

Experiment 1b: Sequential Change Detection (Direct Replication)

Method

Given the importance of direct replications, we reran the same experiment on a larger group of 400 independent observers from the same pool (145 female; mean age = 25.89 years). This preregistered sample size was chosen to be exactly twice that from Experiment 1a and, again, excluded observers (with replacement) according to the same two preregistered criteria (n = 19 and n = 28, respectively).

Results

As depicted in Figure 2c, the results conformed to the same pattern as in Experiment 1a: Changes were again detected much more accurately on underlying-object-change trials (61.50%) compared with cloth-change trials (34.00%), χ² (N = 400) = 30.31, p < .001, effect-size index w = 0.28.

Experiment 1c: Sequential Change Detection (Silhouettes)

The pattern of results observed in Experiments 1a and 1b cannot be explained by brute change magnitude because, by design, there was always greater visual change on cloth-change trials than underlying-object-change trials. But it could be predicted by attention and/or memory prioritizing “deep” object information beyond “superficial” cloth contours. This view also predicts that the observed change-detection patterns should disappear if the cloth/object distinction itself were eliminated in the image. We tested this by presenting stimulus silhouettes, as in Figure 2d.

Method

This experiment was identical to Experiment 1a, except as noted. A new set of 200 observers from the same pool (71 female; mean age = 23.67 years) was recruited, once again excluding observers (with replacement) according to the same two preregistered criteria (n = 6 and n = 24, respectively). Stimuli were identical to those used in Experiment 1a, except that they were presented as silhouettes (with the internal areas filled with a single shade of gray equal to that of the cloth resting directly on the object surface; No. A4A4A4).

Results

As depicted in Figure 2e, change detection did not differ on underlying-object-change trials (68.00%) compared with cloth-change trials (70.00%). χ² (N = 200) = 0.09, p = .760, effect-size index w = 0.02—and this null effect (the slight numerical difference for which was actually in the opposite direction) differed from the robust effect observed in Experiment 1a (odds ratio = 2.28, p < .05).

Experiment 1d: Sequential Change Detection (Silhouettes, Direct Replication)

Method

Given the importance of direct replications, we reran Experiment 1c on a larger group of 400 independent observers from the same pool (160 female; mean age = 26.20 years). This preregistered sample size was chosen to be exactly twice that from Experiment 1c (and equal to that from Experiment 1b), again excluding observers (with replacement) according to the same two preregistered criteria (n = 21 and n = 38, respectively).

Results

As depicted in Figure 2f, change detection did not differ on underlying-object-change trials (65.00%) compared with cloth-change trials (61.50%), χ² (N = 400) = 0.53, p = .468, effect-size index w = 0.04—and this null effect differed from the robust difference observed in Experiment 1b (odds ratio = 2.67, p < .001).

Experiment 2a: Simultaneous Probe Comparison

Performance in the previous experiments seemed to depend not on the image properties themselves but rather on how various contours were represented as a function of intuitive physics—in terms of reflecting deep object contours versus superficial cloth contours. But are these representations formed during online perception or only later—perhaps during retrieval from memory, after the images themselves have disappeared? We addressed this using a paradigm in which the relevant information is always simultaneously visible on the display. Inspired by studies of “same-object (dis)advantages” in object-based attention (e.g., Egly et al., 1994; Marino & Scholl, 2005), we tested this using probe comparison. Two probes appeared atop an image of a cloth-covered object (as depicted in Fig. 3a), and observers simply reported whether they were identical. Both probes always appeared on the images, but we predicted that performance would vary on the basis of whether they both appeared on object regions (vs. one appearing on a cloth region).

Fig. 3.

(a) Depictions of the two key conditions in Experiments 2a and 2b: The two probes could both appear on object regions, or one could appear on a cloth region. (b) Average accuracy for each probe position in Experiment 2a. (c) Average accuracy for each probe position in Experiment 2b. In both graphs, error bars depict 95% confidence intervals after subtracting shared variance. Asterisks indicate significant differences between conditions (**p < .01).

Method

Participants

One hundred observers (33 female; mean age = 24.94 years) participated for monetary compensation using the Prolific online platform (Palan & Schitter, 2018). This preregistered sample size was determined before data collection began.

Stimuli

Eight underlying objects (four base-objects and four corresponding modified-objects) were generated and then covered with cloths as in the previous experiments, except without any constraints involving change magnitudes. Each probe consisted of two 3-pixel gray dots (No. 5E5E5E) that were initially vertically aligned, with their centers 5 pixels apart. During their actual presentation, each probe could then be rotated by 30°, 50°, 310°, or 330°. Two such probes were then placed on each cloth-covered object stimulus during each trial, as described below.

Procedure and design

Using the same apparatus from the previous experiments, we began each trial with the image (of an object covered by a cloth) appearing in a random location. After 350 ms, two probes appeared on the image, and 300 ms later, both the probes and the images disappeared. Observers then reported whether the two probes were identical (i.e., whether the dot pair in each probe had the same degree of rotation)—where nonidentical probe orientations always differed by 20°. On object/object trials, both probes were located on object regions (in one of four fixed image-relative locations). On corresponding object/cloth trials, one probe was located on an object region, whereas the other was located on a cloth region—always equating the interprobe distances and directions across these two trial types, as depicted in Figure 3a (and these equated locations also prevented any differential influence of horizontal vs. vertical arrangements; cf. Z. Chen & Cave, 2019).

Each observer completed 32 trials, presented in a different random order for each observer: 4 base-objects × 2 conditions (object/object vs. object/cloth) × 2 possible base probe rotations (30°/50° or 310°/330°) × 2 probe rotation matching possibilities (identical vs. different). Observers were excluded (with replacement) according to two preregistered criteria. First, in a postexperimental debriefing phase, observers self-reported how well they paid attention (on a continuous scale ranging from 1 = very distracted to 100 = very focused), and we excluded observers who self-reported an attention level below 70 (n = 12). Second, we also excluded observers whose mean accuracy was lower than 60% (who were not already excluded via Criterion 1; n = 35). Individual trials with response times 2 or more standard deviations away from the mean response time of all observers were also excluded (on average, 0.97 trials/observer).

Results

Our preregistered analysis plan made no prediction as to the specific direction of a performance difference across conditions, because past studies have observed both same-object advantages and same-object disadvantages depending on subtle stimulus differences (for a review, see H. Chen & Huang, 2015). (Our underlying theoretical question was only whether the visual system was drawing the cloth/object distinction—and so a reliable, systematic difference in either direction serves to support that possibility.) As depicted in Figure 3b, probe comparison performance was better on object/cloth trials (82.12%) compared with object/object trials (77.54%), t(99) = 3.18, p = .002, d = 0.32.

Experiment 2b: Simultaneous Probe Comparison (Direct Replication)

Method

Given the importance of direct replications, we reran the same experiment with an independent group of 100 observers from the same pool (40 female; mean age = 24.15 years). This preregistered sample size was chosen to match that of Experiment 2a. Using the same preregistered criteria, we excluded observers on the basis of self-reported attention levels (n = 8) and mean accuracy (n = 24), and we excluded trials on the basis of response time variance (on average, 0.73 trials/observer).

Results

As depicted in Figure 3c, probe comparison performance was again better on object/cloth trials (79.16%) compared with object/object trials (75.13%), t(99) = 2.67, p = .009, d = 0.27.

General Discussion

This study, as in so much of vision science, involves a central contrast between images and percepts. The raw images used in this study did not explicitly distinguish between those contours reflective of the deep structure of the underlying (covered) objects and those reflective only of the superficial contours of the (covering) cloths themselves. Yet observers’ percepts clearly respected these distinctions, taking into account the intuitive physics of how cloth, gravity, and objects interact, to prioritize some image contours over others.

This study did not aim just to verify that people are able to draw this distinction, because this has been directly demonstrated in other recent work (Phillips & Fleming, 2020; Ullman et al., 2019; Yildirim et al., 2016). Rather, we explored what types of mental processes are involved, considering the possibility that the cloth/object distinction is drawn during seeing itself, rather than only higher-level reasoning and decision-making. Thus, our experiments differed from prior intuitive physics studies in two key ways. First, instead of posing explicit questions, we employed subtler performance-based measures, which observers cannot intentionally control. Second, whereas all past studies in this domain directly asked observers about the soft materials and/or the underlying objects, this distinction was always completely task irrelevant here.

These themes were apparent in the results from two converging experimental paradigms (both involving high power and direct replications). In change detection (Experiments 1a–1d), observers were better able to detect image changes that reflected different underlying objects versus changes that reflected only superficial cloth contours—even though the latter were more visually extreme in multiple ways. These results (which disappeared when using silhouettes) were especially striking, showing not only that attention and memory were drawing the cloth/object distinction (despite its task irrelevance) but also that observers apparently could not stop this—as they would have performed better had these underlying representations not prioritized the object regions. And in probe comparison (Experiments 2a and 2b), observers’ accuracy when comparing two probes differed depending on probe placement on cloth versus object regions (equating distance and direction)—despite these categories being task irrelevant. These studies thus demonstrate that a facility with this form of intuitive physics occurs to some degree automatically and incidentally, as a part of seeing such stimuli in the first place. Note that this conclusion is orthogonal to most other recent accounts of intuitive physics, such as those that appeal to simulation, or notions of a “game engine” in the mind (e.g., Battaglia et al., 2013; Hamrick et al., 2016; Ullman et al., 2017). From that perspective, what the current results demonstrate is that some such simulations are performed in a relatively automatic (or even irresistible) manner during seeing itself, rather than being triggered only by higher-level goals and intentions.

The online testing platform used in these studies involves a population that is diverse along many dimensions (see Palan & Schitter, 2018), but we cannot generalize the current results beyond this population, and further studies will be required to test whether such effects also occur in people who are not frequent participants in online studies. Similarly, although the stimuli used in the studies were rendered in photorealistic ways, they still used relatively simple “‘Tetris-like’” stimuli under the cloths (in order to allow for precise and systematic changes), and so further research will be necessary in order to generalize such results to other classes of cloth-covered stimuli—as well as to soft materials of varying thicknesses and degrees of stiffness.

Our results can be understood by analogy to the perception of lightness. When raw images are viewed, some image luminance information is highlighted in visual processing because it is informative about the deeper underlying reflectances of objects—but other image luminance information is effectively discounted because it reflects (merely) highly variable details of ambient lighting (for a review, see Adelson, 2000). The current experiments demonstrate the same pattern with cloth physics: When raw images of cloth-covered objects are viewed, some image contours are highlighted in visual processing because they are informative of the deeper underlying object structure—but other image contours are effectively discounted because they reflect (merely) highly variable details of how the cloth was draped. In perception, attention, and memory, visual processing may not only “discount the illuminant” but also “discount the cloth.” And this type of dynamic during visual processing may be just as integral to intuitive physics in our mental lives as is higher-level reasoning about physics.

Supplemental Material

sj-xlsx-1-pss-10.1177_09567976221109194 – Supplemental material for Seeing Soft Materials Draped Over Objects: A Case Study of Intuitive Physics in Perception, Attention, and Memory

Supplemental material, sj-xlsx-1-pss-10.1177_09567976221109194 for Seeing Soft Materials Draped Over Objects: A Case Study of Intuitive Physics in Perception, Attention, and Memory by Kimberly W. Wong, Wenyan Bi, Amir A. Soltani, Ilker Yildirim and Brian J. Scholl in Psychological Science

Footnotes

Acknowledgements

For helpful conversation and/or comments on earlier drafts of this article, we thank the members of both the Yale Perception & Cognition Lab and the Yale Cognitive & Neural Computation Lab.

Transparency

Action Editor: Marc Buehner

Editor: Patricia J. Bauer

Author Contributions

K. W. Wong, I. Yildirim, and B. J. Scholl designed the experiments and wrote the initial manuscript, which was subsequently edited by all the authors. W. Bi and A. A. Soltani developed the methods for stimuli creation and convolutional neural network (CNN) integration. K. W. Wong conducted the experiments and analyzed the data with input from W. Bi, I. Yildirim, and B. J. Scholl. All the authors approved the final manuscript for submission.

ORCID iDs

Kimberly W. Wong

Wenyan Bi

Brian J. Scholl

Supplemental Material

Additional supporting information can be found at

References

Adelson

E. H.

(2000). Lightness perception and lightness illusions. In Gazzaniga

(Ed.), The new cognitive neurosciences (2nd ed., pp. 339–351). MIT Press.

Battaglia

P. W.

Hamrick

J. B.

Tenenbaum

J. B.

(2013). Simulation as an engine of physical scene understanding. Proceedings of the National Academy of Sciences, USA, 110, 18327–18332.

Shah

A. D.

Wong

K. W.

Scholl

B. J.

Yildirim

(2021, July 29). Perception of soft materials relies on physics-based object representations: Behavioral and computational evidence [Paper presentation]. Proceedings of the 43rd Annual Meeting of the Cognitive Science Society, Virtual meeting.

Block

(2022). The border between seeing and thinking. Oxford University Press.

Chen

Huang

(2015). Solving the paradox between same-object advantage and different-object advantage. Vision Research, 115, 128–134.

Chen

Cave

K. R.

(2019). When is object-based attention not based on objects? Journal of Experimental Psychology: Human Perception and Performance, 45, 1062–1082.

de Leeuw

J. R.

(2015). jsPsych: A JavaScript library for creating behavioral experiments in a web browser. Behavior Research Methods, 47, 1–12.

Egly

Driver

Rafal

R. D.

(1994). Shifting visual attention between objects and locations: Evidence from normal and parietal lesion subjects. Journal of Experimental Psychology: General, 123, 161–177.

Firestone

Scholl

B. J.

(2016). Cognition does not affect perception: Evaluating the evidence for “top-down” effects. Behavioral and Brain Sciences, 39, Article e229. https://doi.org/10.1017/S0140525X15000965

10.

Firestone

Scholl

B. J.

(2017). Seeing physics in the blink of an eye. Journal of Vision, 17(10), Article 203. https://doi.org/10.1167/17.10.203

11.

Hamrick

J. B.

Battaglia

P. W.

Griffiths

T. L.

Tenenbaum

J. B.

(2016). Inferring mass in complex scenes by mental simulation. Cognition, 157, 61–76.

12.

Kaiser

M. K.

Proffitt

D. R.

Anderson

(1985). Judgments of natural and anomalous trajectories in the presence and absence of motion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 795–803.

13.

Kubricht

J. R.

Holyoak

K. J.

(2017). Intuitive physics: Current research and controversies. Trends in Cognitive Sciences, 21, 749–759.

14.

Macklin

Müller

Chentanez

Kim

T. Y.

(2014). Unified particle physics for real-time applications. ACM Transactions on Graphics, 33(4), Article 153. https://doi.org/10.1145/2601097.2601152

15.

Marino

A. C.

Scholl

B. J.

(2005). The role of closure in defining the “objects” of object-based attention. Perception & Psychophysics, 67, 1140–1149.

16.

McCloskey

(1983). Intuitive physics. Scientific American, 248, 122–131.

17.

McCloskey

Caramazza

Green

(1980). Curvilinear motion in the absence of external forces: Naïve beliefs about the motion of objects. Science, 210, 1139–1141.

18.

McCloskey

Washburn

Felch

(1983). Intuitive physics: The straight-down belief and its origin. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 636–649.

19.

Palan

Schitter

(2018). Prolific.ac—a subject pool for online experiments. Journal of Behavioral and Experimental Finance, 17, 22–27.

20.

Phillips

Fleming

R. W.

(2020). The Veiled Virgin illustrates visual segmentation of shape by cause. Proceedings of the National Academy of Sciences, USA, 117, 11735–11743.

21.

Shepard

R. N.

(1994). Perceptual-cognitive universals as reflections of the world. Psychonomic Bulletin & Review, 1, 2–28.

22.

Simonyan

Zisserman

(2015, May 18). Very deep convolutional networks for large-scale image recognition [Conference session]. The 3rd International Conference on Learning Representations (ICLR 2015), San Diego, CA.

23.

Tse

P. U.

(1999). Volume completion. Cognitive Psychology, 39, 37–68.

24.

Ullman

T. D.

Kosoy

Yildirim

Soltani

A. A.

Siegel

M. H.

Tenenbaum

Spelke

E. S.

(2019, July 25). Draping an elephant: Uncovering children’s reasoning about cloth-covered objects [Paper presentation]. Proceedings of the 41st Annual Conference of the Cognitive Science Society, Montreal, Canada.

25.

Ullman

T. D.

Spelke

E. S.

Battaglia

Tenenbaum

J. B.

(2017). Mind games: Game engines as an architecture for intuitive physics. Trends in Cognitive Sciences, 21, 649–665.

26.

Yildirim

Siegel

Soltani

A. A.

Chaudhuri

Tenenbaum

J. B.

(2022). Seeing objects under cloth: Evidence for physics-based models and analysis-by-synthesis in 3D shape perception. Manuscript submitted for publication.

27.

Yildirim

Siegel

Tenenbaum

J. B.

(2016, August 11). Perceiving fully occluded objects via physical simulation [Paper presentation]. Proceedings of the 38th Annual Conference of the Cognitive Science Society, Philadelphia, PA.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.68 MB