Pith. sign in

REVIEW 4 major objections 5 minor 38 references

A Software to Detect OCC Emotion, Big-Five Personality and Hofstede Cultural Dimensions of Pedestrians from Video Sequences

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that the GeoMind software can estimate pedestrian Big-Five personality, OCC emotions, and Hofstede cultural dimensions from motion and grouping features in video.

desk verdict A functional feature-extraction and visualization tool, but the 'detection' of personality, emotion, and culture is unvalidated hand-crafted mapping. read the letter →

arxiv 1908.06484 v1 pith:N6PST5FY submitted 2019-08-18 cs.CV cs.HC

Rodolfo Migon Favaretto , Victor Araujo , Soraia Raupp Musse , Felipe Vilanova , Angelo Brandelli Costa This is my paper · ORCID
classification cs.CVcs.HC
keywords crowdanalysispersonalitydetectionBig-FiveOCEANOCCemotionmodelHofstedeculturaldimensionspedestriantrackingvideoGeoMind
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

GeoMind is a software application that takes pedestrian tracking data from video and converts movement patterns into three layers of psychological description: Big-Five personality traits, OCC emotions, and Hofstede cultural dimensions. The authors' central hypothesis is that intrinsic human variables show up in how people move, cluster, and keep distance in space and time. To test this, they hand-built equations linking physical and social features such as speed, angular variation, collectivity, and isolation to selected NEO PI-R items, then used an existing mapping from personality to emotions and a group-based mapping to cultural dimensions. They report promising results on videos from different countries and provide a viewer for inspecting the estimated traits. If the premise holds, the software offers a way to read psychological and cultural information from ordinary crowd footage without questionnaires.

What carries the argument

The central mechanism is a multi-stage mapping. Low-level features per pedestrian are speed, angular variation, isolation, socialization, and collectivity; collectivity is computed as a decaying Gaussian sum over pairwise differences in speed and orientation. A neural network trained with proxemics inputs turns these into a socialization score. Personality is obtained by hand-built equations linking 25 NEO PI-R items to the features, emotion follows from a fixed OCEAN-to-OCC table, and cultural dimensions are computed from group size, mean group distance, group angular variation, cohesion, collectivity, and speed. Each stage is deterministic once the tracking file is supplied, so the whole pipeline is a software tool that turns trajectories into trait estimates.

What would settle it

Record crowd videos in several countries with known published Hofstede scores, run GeoMind on them, and compare the predicted country-level dimensions with the published indices; if the predicted ordering does not match the published ordering across countries, the cultural mapping is not detecting culture.

Watch

Extended reading notes

Core claim

The paper's central claim is that observable pedestrian dynamics carry enough information to reconstruct OCEAN personality, OCC emotions, and Hofstede cultural dimensions. For each tracked person the software averages per-frame features into a vector, maps selected NEO PI-R items to these features through empirically defined equations, converts the resulting personality scores into OCC emotion scores through a fixed OCEAN-to-OCC table, and aggregates group-level statistics into Hofstede dimensions such as individualism/collectivism, power distance, long-term orientation, masculinity, and indulgence. The authors present this as a complete detection pipeline, not just a visualization, and they report that tests on the Cultural Crowds dataset give very satisfactory results.

Load-bearing premise

The load-bearing premise is that measurable motion cues like speed, spacing, and grouping actually reflect inner personality, emotion, and cultural traits, so the hand-built equations are measuring something real.

Editorial extensions

If this is right

  • With only tracking files as input, the tool produces per-pedestrian Big-Five scores, OCC emotions, and per-video Hofstede dimensions, making psychological profiling possible on existing surveillance or public footage.
  • The cultural mapping gives a video-based route to cross-cultural comparison: two crowd videos from different countries can be compared on the same Hofstede scales without running questionnaires.
  • The outputs can feed crowd simulations and games, giving virtual characters personalities, emotions, and cultural traits derived from real pedestrian behavior.
  • The viewer lets researchers inspect the estimated traits per agent over time, which supports qualitative checking of whether the detections make sense in context.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct test the authors do not run: compare GeoMind's country-level Hofstede outputs to Hofstede's published national scores; a strong positive correlation would be independent evidence that the cultural equations capture real cultural variation.
  • Because the emotion stage is a fixed table applied to estimated personality, errors in the personality equations propagate into every emotion; an ablation that bypasses personality and predicts emotions directly from features could show how much of the emotion signal is real.
  • The hand-built NEO PI-R equations could be replaced by a supervised regressor trained on questionnaire scores paired with tracking data, turning the claimed detection into a measurable predictive model.
  • The authors treat all five OCEAN traits as individually estimable per pedestrian, but the underlying features are mostly about movement and grouping; a skeptical extension would test whether the five estimated dimensions are statistically distinguishable or collapse into one or two motion factors.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents GeoMind, a Matlab-based software application that takes pedestrian tracking data from video sequences, extracts features such as speed, angular variation, collectivity, socialization, and isolation, and then maps those features to Big-Five (OCEAN) personality scores, OCC emotion scores, and Hofstede cultural dimensions. A Unity3D viewer is also described for visualizing the extracted features and computed attributes. The authors claim that the software 'detects' personality, emotion, and cultural aspects of pedestrians, based on the hypothesis that intrinsic human variables are related to how people behave in space and time. The paper describes the pipeline, the input/output formats, the graphical interface, and several qualitative example results, but it does not report any quantitative validation against external ground truth.

Significance. If the central claim were established, GeoMind would be a notable tool for crowd analysis, with potential applications in surveillance, serious games, and social-science research. The paper's strengths are the public availability of the software, the concrete description of a complete pipeline from tracking files to visualizations, and the integration of three psychological/cultural frameworks in one system. However, the significance is conditional: the manuscript provides no evidence that the computed OCEAN, OCC, and Hofstede values correspond to the actual psychological or cultural states of the filmed pedestrians. As submitted, the paper is a software description with illustrative outputs rather than a validated detection method.

major comments (4)
  1. [Section 2.2] The personality module is the load-bearing part of the pipeline, but its core equations are asserted without validation. For example, Q1 = s_i + 1/alpha_i combines quantities with different units without normalization and is undefined when alpha_i = 0. More importantly, no evidence is presented that any of the 25 equations correspond to NEO PI-R item responses: there is no item-level or factor-level comparison to questionnaire scores, no correlation analysis, and no inter-rater agreement measure. The resulting OCEAN scores are therefore not established as measurements of the Big-Five constructs; they are computed features relabeled as personality dimensions.
  2. [Section 2.3, Table 1] The OCC emotion scores are deterministic functions of the OCEAN scores through a hand-assigned sign table (with OCEAN values thresholded at 0.5). Because the OCEAN scores themselves lack external validation, and because no emotion ground truth (self-report, facial expression annotation, or behavioral criterion) is used anywhere in the paper, the emotion labels produced by GeoMind cannot be considered detections of anger, fear, happiness, or sadness.
  3. [Section 2.4] The Hofstede dimension equations reduce to the authors' own definitions: collectivism is defined as the percentage of grouped pedestrians, individualism as the percentage of lonely pedestrians, and MAS and IND are weighted combinations with hand-chosen weights sigma1 = 0.5 and rho1 = 0.5. The paper reports no comparison to Hofstede's published country-level scores or any other external criterion. The statement in Section 5 that results are 'very satisfactory' is not supported by any quantitative evaluation, so the cultural-detection claim is unsupported.
  4. [Sections 3.2 and 5] The evaluation section consists of screenshots, descriptive comments about the graphical interface, and the assertion that results are very satisfactory. There are no error bars, no baselines, no statistical tests, no cross-validation, and no comparison against ground-truth labels for personality, emotion, or culture. Given that the central contribution is the detection of latent psychological constructs, the absence of any quantitative validation is a load-bearing gap, not a presentation issue.
minor comments (5)
  1. [Section 2.1, Eq. (1)] The constants beta = 0.3 and w1 = w2 = 1 are described as empirically defined, but no empirical procedure is reported. Since these constants directly control the collectivity values, a sensitivity analysis or at least a justification of the chosen values is needed.
  2. [Section 2.1] The description of the socialization ANN is incomplete: the paper reports 10 hidden layers and 96% accuracy on 16,000 samples with a 70/30 split, but does not specify the number of neurons per layer, whether the 96% is on the training or validation portion, or how the ground-truth labels for socialization were obtained. This prevents reproduction of the socialization module.
  3. [Section 2.2, Eq. for Q1] The equation Q1 = s_i + 1/alpha_i is undefined when alpha_i = 0, which can occur when a pedestrian moves in a perfectly straight line. The paper should specify how zero angular variation is handled, and should also clarify how s_i and 1/alpha_i are normalized before summation.
  4. [Figure 6 caption] The caption says 'From (g) to (f)' where it should say '(g) to (j)'. There are also several typographical errors elsewhere, such as 'App Desinger' instead of 'App Designer' in Section 3.
  5. [Section 6 and website references] The paper states that supporting data are available on request and that the software is downloadable from a personal website. For reproducibility, the dataset and code should be deposited in a permanent archival repository with versioning.

Circularity Check

3 steps flagged · score 7.0 of 10

The personality, emotion, and cultural outputs are defined by the paper's own feature equations and self-citations, so the central detection claim reduces to those inputs by construction.

  1. self definitional [Section 2.4, Cultural Aspects (COL/IDV definitions, MAS and IND equations)]
    "Indeed, collectivism (COL) is a % of people grouped, while the individualism (IDV) is a % of lonely people. ... MAS = σ1GCk + (1 −σ1)LTO, where σ1 = 0.5 is the empirically chosen weight. ... IND = ρ1Sk + (1 −ρ1)COL, where ρ1 = 0.5 is an empirically chosen weight."

    The Hofstede dimensions are not validated against any external cultural criterion. By the paper's own text, IDV and COL are literally defined as the percentage of lonely and grouped pedestrians, so reporting them as detected cultural dimensions is just renaming the group statistics. MAS and IND are similarly defined as weighted sums of already-computed features (group cohesion, LTO, speed, collectivity) with hand-picked weights. Every cultural output is therefore equivalent to the input features under new names.

  2. fitted input called prediction [Section 2.2, Personality (item equations and Q1)]
    "As we describe in details in [15], we proposed a series of empirically defined equations to map pedestrian features to OCEAN dimensions. ... In order to answer the items with data coming from real video sequences, we propose equations that could represent each one of the 25 items with features extracted from videos. ... So the equation for this item is Q1 = s_i + 1/α_i."

    No NEO PI-R responses, questionnaire scores, or other personality ground truth are used. The OCEAN outputs are constructed directly from hand-written formulas, e.g. speed plus inverse angular variation, which the authors choose to stand for an item's answer. Thus the detected personality score is not an estimate of an external construct but a deterministic relabeling of the same motion features; any apparent fit is the formula itself, not empirical evidence.

1 more flagged steps
  1. self citation load bearing [Section 2.3, Emotion (Table 1 and citation [17])]
    "As we presented in [17], we proposed a way to map the OCEAN dimensions of each pedestrian in OCC Emotion model. This mapping is described in Table 1. ... if an individual has a high value for Extraversion (for example, E = 0.9), following the mapping in Table 1, this individual can present signals of happiness (i.e. If E+ then Happiness= 1) and should not be angry (i.e. If E+ then Anger= -1)."

    The stated emotion detection is a sign-multiplication of the previously constructed OCEAN values using Table 1, and Table 1 is introduced only by citation to the authors' own prior work [17]. No emotion labels, expressions, or independent affective measurements enter. The emotion scores therefore inherit the by-construction character of the personality mapping: they are deterministic transformations of the same feature-derived inputs, with the load-bearing mapping supplied by self-citation rather than external validation.

full rationale

The abstract and introduction claim that GeoMind detects OCEAN personality, OCC emotions, and Hofstede cultural dimensions from pedestrian trajectories. Walking the derivation chain, every one of those outputs is a direct product of the paper's own hand-authored equations: cultural dimensions are group percentages and weighted feature means, personality items are arbitrary functions of speed and angular variation, and emotions are sign tables applied to those OCEAN values. No external criterion, questionnaire validation, ground-truth label, or statistical baseline is introduced; the statement in Section 5 that results are very satisfactory is an unbenchmarked self-assessment. The central detection claim is therefore circular: the outputs reduce by construction to the tracked motion features plus the authors' chosen weights. I do not penalize the self-citations themselves, but here they are load-bearing because the cited prior work is the only source of the personality and emotion mappings and is not independently verified.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on multiple hand-fitted parameters and strong domain assumptions about the link between visible motion and invisible psychological traits. All cultural and emotional outputs are computed from equations defined by the authors; no independent construct validation is provided.

free parameters (6)
  • Collectivity decay constant beta = 0.3
    Empirically defined decay constant in Eq. 1; chosen without stated basis.
  • Collectivity weights w1, w2 = w1=1, w2=1
    Constants in Eq. 1 that regulate speed and orientation offset; set to 1 with no justification.
  • MAS weight sigma1 = 0.5
    Empirically chosen weight in MAS = sigma1*GC + (1-sigma1)*LTO in Section 2.4.
  • IND weight rho1 = 0.5
    Empirically chosen weight in IND = rho1*S + (1-rho1)*COL in Section 2.4.
  • OCEAN factor positive/negative threshold = 0.5
    Threshold dividing O+ from O- in the emotion mapping in Table 1; changes which emotions are activated.
  • 25 NEO PI-R item scoring equations = defined in Ref. [15]
    Hand-designed mapping from video features to item answers; the central personality measurement tool is asserted in Section 2.2 but all details are deferred to a prior paper.
assumptions (4)
  • domain assumption Intrinsic human variables such as emotion, personality, and culture are systematically related to pedestrian spatial-temporal behavior.
    Stated as the main hypothesis in the abstract and Section 2; the whole pipeline assumes this relationship without psychological validation.
  • domain assumption The 25 selected NEO PI-R items can be answered by geometric video features via hand-written equations.
    Section 2.2 selects items and proposes equations such as Q1 = s_i + 1/alpha_i; no evidence connects these equations to actual questionnaire responses.
  • domain assumption Hofstede cultural dimensions can be operationalized as simple group statistics such as IDV as the percentage of lonely people and PDI as mean group distance.
    Section 2.4 defines the dimensions directly from group features; this operationalization is asserted and not validated cross-culturally.
  • domain assumption The Saifi et al. OCEAN-to-OCC mapping is valid for individuals whose OCEAN scores come from video-derived features.
    Section 2.3 applies the mapping table without checking its applicability to this new measurement context.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Software to Detect OCC Emotion, Big-Five Personality and Hofstede Cultural Dimensions of Pedestrians from Video Sequences." pith.science (2026). https://pith.science/paper/N6PST5FY

@misc{pith2026190806484,
  author       = {Pith},
  title        = {Pith review of: A Software to Detect OCC Emotion, Big-Five Personality and Hofstede Cultural Dimensions of Pedestrians from Video Sequences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N6PST5FY}},
  note         = {Machine review of arXiv:1908.06484}
}
read the original abstract

This paper presents a video analysis application to detect personality, emotion and cultural aspects from pedestrians in video sequences, along with a visualizer of features. The proposed model considers a series of characteristics of the pedestrians and the crowd, such as number and size of groups, distances, speeds, among others, and performs the mapping of these characteristics in personalities, emotions and cultural aspects, considering the Cultural Dimensions of Hofstede (HCD), the Big-Five Personality Model (OCEAN) and the OCC Emotional Model. The main hypothesis is that there is a relationship between so-called intrinsic human variables (such as emotion) and the way people behave in space and time. The software was tested in a set of videos from different countries and results seem promising in order to identify these three different levels of psychological traits in the filmed sequences. In addition, the data of the people present in the videos can be seen in a crowd viewer.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages

  1. [32]

    L. Saifi, A. Boubetra, and F. Nouioua. An approach for emotions and behavior modeling in a crowd in the presence of rare events. AB, 24(6), 2016

    work page 2016

  2. [1]

    A. E. Abele and B. Wojciszke. Agency and communion from the perspective of self versus others. JPSP, 93(5):751–763, 2007

    work page 2007

  3. [2]

    How much do you perceive this? An analysis on perceptions of geometric features, personalities and emotions in virtual humans (Extended Version)

    V . Araujo, R. M. Favaretto, P. Knob, S. R. Musse, F. Vilanova, and A. B. Costa. How much do you perceive this? an analysis on perceptions of geometric features, personalities and emotions in virtual humans (extended version). arXiv preprint arXiv:1904.11084, 2019

    work page Pith review arXiv 1904

  4. [3]

    Araujo, R

    V . Araujo, R. Migon Favaretto, P. Knob, S. Raupp Musse, F. Vilanova, and A. Brandelli Costa. How much do you perceive this?: An analysis on perceptions of geometric features, personalities and emotions in virtual humans. In Proceedings of the 19th ACM International Conference on Intelligent Virtual Agents , IV A ’19, pages 179–181, New York, NY , USA, 2019. ACM

    work page 2019

  5. [4]

    G. L. C. S. Hall and J. B. Campbell. Theories Of Personality. John Wiley & Sons, New Jersey, fourth edition, 1998

    work page 1998

  6. [5]

    Z. Cai, Z. L. Yu, H. Liu, and K. Zhang. Counting people in crowded scenes by video analyzing. In 9th IEEE ICIEA, pages 1841–1845, June 2014. 11

    work page 2014

  7. [6]

    A. B. Chan and N. Vasconcelos. Bayesian poisson regression for crowd counting. In IEEE 12th ICCV, pages 545–551, Sep. 2009

    work page 2009

  8. [7]

    Chandran, L

    A. Chandran, L. A. Poh, and P. Vadakkepat. Identifying social groups in pedestrian crowd videos. In ICAPR, pages 1–6, Jan 2015

    work page 2015

Show all 38 references
  1. [8]

    Chattaraj, A

    U. Chattaraj, A. Seyfried, and P. Chakroborty. Comparison of pedestrian fundamental diagram across cul- tures. ACS, 12(03):393–405, 2009

    2009

  2. [9]

    Costa and R

    P. Costa and R. McCrae. Revised NEO Personality Inventory (NEO PI-R) and NEO Five-Factor Inventory (NEO-FFI). PAR, 1992

    1992

  3. [10]

    L. Dihl, E. S. Testa, P. Knob, G. L. da Silva, R. M. Favaretto, M. F. de Alcântara, and S. R. Musse. Gen- erating cultural characters based on hofstede dimensions. In Virtual Humans and Crowds for Immersive Environments (VHCIE), 2017 IEEE, pages 1–5. IEEE, 2017

    2017

  4. [11]

    Q. Dream. Detroit. Available at: https://www.playstation.com/pt-br/games/detroit-become-human-ps4/, 2019

    2019

  5. [12]

    Durupınar, U

    F. Durupınar, U. Güdükbay, A. Aman, and N. I. Badler. Psychological parameters for crowd simulation: From audiences to mobs. IEEE transactions on visualization and computer graphics, 22(9):2145–2159, 2016

    2016

  6. [13]

    Ekman and W

    P. Ekman and W. V . Friesen. Constants across cultures in the face and emotion. JPSP, 17(2), 1971

    1971

  7. [14]

    R. M. Favaretto, L. Dihl, R. Barreto, and S. R. Musse. Using group behaviors to detect hofstede cultural dimensions. In 2016 IEEE International Conference on Image Processing (ICIP) , pages 2936–2940, Sep. 2016

    2016

  8. [15]

    R. M. Favaretto, L. Dihl, S. R. Musse, F. Vilanova, and A. B. Costa. Using big five personality model to detect cultural aspects in crowds. In 2017 30th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI), pages 223–229, Oct 2017

    2017

  9. [16]

    R. M. Favaretto, L. L. Dihl, and S. R. Musse. Detecting crowd features in video sequences. In 2016 29th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI), pages 201–208, Oct 2016

    2016

  10. [17]

    R. M. Favaretto, P. Knob, S. R. Musse, F. Vilanova, and Â. B. Costa. Detecting personality and emotion traits in crowds from video sequences. Machine Vision and Applications, Oct 2018

    2018

  11. [18]

    Feng and B

    L. Feng and B. Bhanu. Understanding dynamic social grouping behaviors of pedestrians. IEEE JSTP , 9(2):317–329, March 2015

    2015

  12. [19]

    Fridman, G

    N. Fridman, G. A. Kaminka, and A. Zilka. The impact of culture on crowd dynamics: An empirical approach. In AAMAS, pages 143–150, Richland, SC, 2013

    2013

  13. [20]

    W. G., R. T. C., and B. R. Vision-based analysis of small groups in pedestrian crowds. IEEE PAMI, 34(5):1003–1016, 2012

    2012

  14. [21]

    L. R. Goldberg. From ace to zombie: Some explorations in the language of personality. In APA, chapter 6, pages 203–234. Minnesota, 1982

    1982

  15. [22]

    description of Personality

    L. R. Goldberg. An alternative "description of Personality": the Big-Five factor structure. JPSP, 59(6):1216– 1229, 1990

    1990

  16. [23]

    Gorbova, I

    J. Gorbova, I. Lüsi, A. Litvin, and G. Anbarjafari. Automated screening of job candidate based on multimodal video processing. In CVPRW, July 2017

    2017

  17. [24]

    Hofstede

    G. Hofstede. Culture’s consequences: comparing values, behaviors, institutions, and organizations across nations. Sage Publications, 2001. 12

    2001

  18. [25]

    Hofstede

    G. Hofstede. Dimensionalizing cultures: The hofstede model in context. Online readings in psychology and culture, 2(1):8, 2011

    2011

  19. [26]

    P. Knob, V . F. de Andrade Araujo, R. M. Favaretto, and S. R. Musse. Visualization of interactions in crowd simulation and video sequences

  20. [27]

    D. Lala, S. Thovuttikul, and T. Nishida. Towards a virtual environment for capturing behavior in cultural crowds. In 6th ICDIM, pages 310–315, Sept 2011

    2011

  21. [28]

    R. R. McCrae and O. P. John. An introduction to the five-factor model and its applications. Journal of Personality, 60(2):175–215, 1992

    1992

  22. [29]

    Migon Favaretto, R

    R. Migon Favaretto, R. Rosa dos Santos, S. Raupp Musse, F. Vilanova, and A. Brandelli Costa. Investigating cultural aspects in the fundamental diagram using convolutional neural networks and virtual agent simulation. Computer Animation and Virtual Worlds, 30(3-4):e1899, 2019

    2019

  23. [30]

    Montag and J

    C. Montag and J. Panksepp. Primary emotional systems and personality: An evolutionary perspective. FP, 8:464, 2017

    2017

  24. [31]

    Ortony, G

    A. Ortony, G. L. Clore, and A. Collins. The cognitive structure of emotions. Cambridge univ., 1990

    1990

  25. [33]

    R. J. Sethi. Towards defining groups and crowds in video using the atomic group actions dataset. In ICIP, pages 2925–2929, 2015

    2015

  26. [34]

    J. Shao, C. Loy, and X. Wang. Scene-independent group profiling in crowd. In IEEE CVPR, pages 2227– 2234, June 2014

    2014

  27. [35]

    Solera, S

    F. Solera, S. Calderara, and R. Cucchiara. Structured learning for detection of social groups in crowd. In 2013 10th IEEE ICAVSBS, pages 7–12, Aug 2013

    2013

  28. [36]

    Solmaz, B

    B. Solmaz, B. E. Moore, and M. Shah. Identifying behaviors in crowd scenes using stability analysis for dynamical systems. TPAMI, 34(10), Oct 2012

    2012

  29. [37]

    F. Yang, J. Shabo, A. Qureshi, and C. Peters. Do you see groups?: The impact of crowd density and viewpoint on the perception of groups. In Proceedings of the 18th International Conference on Intelligent Virtual Agents, pages 313–318. ACM, 2018

    2018

  30. [38]

    B. Zhou, X. Tang, H. Zhang, and X. Wang. Measuring crowd collectiveness. IEEE PAMI, 36(8):1586–1599, Aug 2014. 13

    2014

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.