Pith. sign in

REVIEW 3 major objections 5 minor 37 references

On the development of an AI performance and behavioural measures for teaching and classroom management

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that AI vision and speech analysis can turn 69 hours of classroom recordings into quantifiable teacher-behavior measures, and that teacher educators welcome the resulting dashboard as a non-judgmental self-review tool.

desk verdict A useful, honestly-scoped proof-of-concept for AI-assisted teacher self-review with a new Singapore classroom dataset; the action-recognition validation is thinner than the 'objective snapshot' framing implies, but the paper is worth refereeing. read the letter →

arxiv 2506.11143 v2 pith:JD3E5LMP submitted 2025-06-11 cs.CV

Andreea I. Niculescu , Jochen Ehnes , Chen Yi , Du Jiawei , Tay Chiat Pin , Joey Tianyi Zhou , Vigneshwaran Subbaraju , Teh Kah Kuan
show 4 more authors
Tran Huy Dat John Komar Gi Soong Chee Kenneth Kwok
This is my paper · ORCID
classification cs.CV
keywords teacherbehavioranalysismultimodalclassroomsensingcomputervisionineducationspeechandprosodyteachingreviewdashboardnon-verbalmeasurementSingaporeclassroomseducationalanalytics
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

The paper reports a two-year effort to turn raw classroom recordings into quantifiable measures of teaching behavior. Using synchronized cameras, microphones, and thermal sensors in a Singapore university classroom, the authors tracked where the teacher moves, what the teacher does at the board, and how the teacher's voice sounds, and assembled the results into a proof-of-concept Teacher Trainer dashboard for self-review. The central claim is that a fully automated pipeline can extract these measures without using any verbal content: computer vision supplies movement and gesture, speech analysis supplies vocal delivery, and the dashboard presents both as neutral summaries rather than performance ratings. An informal evaluation with eight teacher educators found the tool clear, usable, and non-judgmental, suggesting that AI-based observation could reduce the manual workload of classroom review while supporting teacher reflection. The paper also contributes the underlying dataset and behavioral measures as a culturally grounded resource for Asian educational contexts.

What carries the argument

The load-bearing machinery is a multimodal sensing-to-measure pipeline followed by a review dashboard. Position tracking couples YOLOv8 person detection with DeepOCSORT re-identification, plus custom fix-ups (teacher position history, a student ID list to prevent reassignment, and exit detection); action recognition runs a pretrained ST-GCN skeleton model over the tracked pose, with slide changes detected by the Segment Anything family and hand gestures by mmpose-extracted hand coordinates. Speech analysis uses Praat to extract loudness, pitch, formants, and prosody, grouped into statistical, contextual, and linguistic features and scored on a normalized [0,100] scale by a Multiple Criteria Decision Making model. The dashboard renders the output: teaching-action proportions, an active-versus-passive teaching-style balance, speaking-style comparisons against literature norms (140 words per minute, Speech Transmission Index 0.75, monotony norm 1.0), a movement heatmap, and time-synchronized review panels. The machinery converts unstructured video and audio into structured, comparable numbers that a teacher can reflect on without re-watching hours of footage.

What would settle it

Hold out a subset of the 45 recordings, have expert observers annotate every teacher action and speech segment, and compare the annotations against the pipeline's outputs: if per-class agreement on pointing at the board stays near the reported 65.06% or if measured speaking rates systematically disagree with expert counts on the same clips, the dashboard's behavioral proportions would not survive independent checking.

Watch

Extended reading notes

Core claim

The paper's central claim is stated in Section 3.2: "We developed quantifiable measures of teacher behaviors using AI-driven computer vision and speech analysis based on sensor data." Concretely, the authors claim that a pipeline of off-the-shelf and pretrained models — YOLOv8 for locating the teacher, DeepOCSORT for keeping the teacher's identity across frames, ST-GCN for recognizing board-writing, pointing, and gesturing, SAM with MobileSAM for detecting slide changes, and mmpose for hand-waving — can label teaching actions with the reported accuracies of 83.33% for writing on the board, 65.06% for pointing, and 88.57% for gesturing, while Praat-based acoustic analysis yields speaking-rate, clarity, and monotony measures benchmarked against published norms. These measures feed the Teacher Trainer dashboard with a summary screen and a synchronized review screen. The paper states explicitly that the current version assigns no performance ratings; its contribution is an objective snapshot of in-class interactions that lets teachers recognize and improve their instructional strategies, and an initial evaluation by eight researchers from the teacher-education side received it as a non-judgmental automated alternative to manual observation.

Load-bearing premise

The load-bearing premise is that pretrained computer-vision models and published speech norms transfer to Singapore classrooms without local fine-tuning or validation; if they misdetect actions or mis-fit local speech, the dashboard's summaries misrepresent the teaching that actually happened.

Editorial extensions

If this is right

  • Teacher educators can review a lesson from the dashboard's summaries instead of manually coding video, reducing the time cost of classroom observation.
  • Because the measures are automated and the tool gives no performance ratings, teachers get a non-judgmental snapshot that encourages constructive self-reflection rather than defensive evaluation.
  • The framework is directly extensible to student-side metrics such as engagement and cognitive load, since the recordings already include student video, thermal images, gaze, and heart-rate signals.
  • Adding higher-order instructional markers — questioning, prompting, encouraging participation — would let the tool capture instructional intent, a gap the evaluators themselves flagged.
  • Content-level analysis of what the teacher says remains out of scope until ethics approval, so the current measures deliberately cover only non-verbal and paralinguistic cues.

Reading between the lines

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

  • The whole argument inherits a transfer assumption: because the vision models and speech norms were not re-validated on local data, the measures would mislead precisely when pretrained models misfire in this classroom (for instance, pointing at the board at the reported 65.06% accuracy) or when the 140-wpm and STI norms do not match Singaporean English delivery.
  • If the pipeline were validated against expert observation on a held-out subset of the 45 recordings, the same measures could plausibly become automated formative feedback at scale, including a lightweight coaching loop that flags changes across a teacher's own lessons.
  • The dataset's unused channels — student gaze, facial blendshapes, thermal imagery, heart rate — suggest a natural next experiment: correlating the teacher-behavior measures with student-attention signals to test whether the dashboard's snapshot actually tracks engagement.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper reports a two-year project at NIE/NTU Singapore that collected a multimodal audio-visual corpus of 45 classroom recordings (about 69 hours) from 12 teachers, annotated low-level teaching actions, and used off-the-shelf computer vision and speech tools to produce quantitative behavioral measures. These measures feed a proof-of-concept 'Teacher Trainer' dashboard with summary and review screens. An informal evaluation with eight NIE researchers is reported as positive. The paper claims that the system provides an 'objective snapshot' of in-class interactions and contributes a culturally grounded methodology for Asian educational contexts, while explicitly noting that it does not assign performance ratings.

Significance. If the measures are valid, the paper would make a useful applied contribution: a classroom dataset in an under-represented Asian context, a reproducible multimodal pipeline using YOLOv8/DeepOCSORT/ST-GCN/SAM and Praat, and a dashboard that teacher educators found usable and non-judgmental. The authors are appropriately cautious in places: Section 3.3.3 concedes the need for a standardized evaluation, and Section 4 explicitly says no performance ratings are assigned. I also find no circularity: the speech norms and MCDM weights are not fitted to a target outcome, and the behavioral scores are summaries of detections. However, the central 'objective snapshot' claim rests on validation evidence that is currently missing, so the significance is conditional on the concerns below being addressed.

major comments (3)
  1. [§3.2.1] The three action-recognition accuracies (83.33% for writing on the board, 65.06% for pointing at the board/screen, and 88.57% for gesturing at the board) are reported without any evaluation protocol: no train/test split, no sample size, no confusion matrix, and no confidence intervals. Section 2.4.2 states that ten annotated videos totaling 13h42m were used 'for use in training the AI model,' but no fine-tuning procedure is described, and the action classes are not standard pretrained action-recognition categories. If the accuracies were computed on the same small annotated set used for training, they are likely optimistic and unrepresentative of the Singapore classroom. Because the Teaching Actions panel in Section 3.3.1 is built directly on these outputs, this is load-bearing for the paper's central claim. Please specify the evaluation protocol, report per-class performance with uncertainty, and, if possible, validate on a small held-out set.
  2. [§3.2.2] The speech measures rely on literature-based norms (140 wpm, STI 0.75, monotony 1.0) that are not validated for the Singapore teaching context, despite the abstract's claim of a 'culturally grounded methodology.' Section 3.2.2 itself notes that British-like clear speech is preferred in Singapore, but no local calibration or comparison is reported. In addition, the mapping from raw Praat features to the normalized [0,100] MCDM scores is not described, so it is unclear what the Speaking Style panel means numerically. Without at least a small validation against expert judgment or locally collected norms, the 'objective snapshot' wording overstates what the speech measures establish.
  3. [§3.3.3] The evaluation with eight NIE researchers is informal, with no questionnaire, sampling rationale, or analysis procedure; the authors themselves state that 'a more standardized evaluation protocol is needed.' This is acceptable for a proof-of-concept, but the abstract and conclusion present the positive feedback as a highlighted outcome of the system. Please either soften the claim (e.g., 'preliminary feedback suggests') or add a structured coding scheme, inter-rater agreement, or a small survey instrument so that the evaluation evidence matches the strength of the claim.
minor comments (5)
  1. [Title] The title is grammatically awkward: 'an AI performance and behavioural measures' should be revised, for example, to 'AI-based performance and behavioural measures for teaching and classroom management.'
  2. [§2.1] The recording arithmetic is hard to follow: 53 recordings and 90 hours are mentioned, then 8 sessions are excluded and 5 pilot sessions are described. Please state explicitly how the 5 pilots and 8 excluded sessions relate to the final 45 recordings and 69 hours.
  3. [§3.2.1] There is a typo in 'pretrained (mmpose weights' — a closing parenthesis is missing. The sentence should read 'pretrained mmpose weights.'
  4. [§3.2.2 and Reference [3]] The monotony norm of 1.0 and the thresholds 0.4 and 1.6 are sourced from a commercial blog (iMotions) rather than a peer-reviewed source; please replace this with a scholarly reference or present the values as provisional working assumptions.
  5. [Data collection information] The dataset is described as available 'upon request' with only an email address; if the venue allows, please add a short data-sharing statement specifying access conditions, anonymization, and any ethical restrictions.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the behavioral measures are descriptive summaries of detections and externally cited norms, not fitted to the target measures.

full rationale

The paper's central claim is that AI-based computer vision and speech analysis produce quantifiable measures of teacher behavior (Section 3.2). Walking the derivation chain, no step reduces to its own input. The action-recognition accuracies (83.33%, 65.06%, 88.57%) are reported for pretrained/fine-tuned ST-GCN/MobileSAM/mmpose models; even though the evaluation protocol is under-specified, the paper does not state that these accuracies were computed on the training data or that the dashboard measures were fitted to the reported outcomes. The speech norms (140 wpm, STI 0.75, monotony 1.0) come from external literature [29, 28, 16, 3], not from the authors' fitted values. The MCDM scoring uses equal weights or reciprocal standard-deviation weights [33] and is explicitly a descriptive normalization, not a learned predictor. The 'Teaching Style' panel is stated to come from manual annotations, not from the AI pipeline. The only self-referential element is the microphone-setup experience cited from the authors' prior acoustic work in Section 2.3; this is contextual engineering background and does not constrain any behavioral measure or dashboard outcome. The informal evaluation by eight NIE researchers is an opinion study, not a circular validation. Concerns about transferability of pretrained models or lack of confidence intervals are correctness/robustness issues, not circularity, because they do not show that any reported 'prediction' is equivalent by construction to a fitted input or a self-citation chain.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the transferability of pretrained models, the proxy validity of low-level actions for teaching behavior, the applicability of literature norms to Singapore, and the representativeness of a small informal evaluation. No new physical entities or fitted parameters are introduced; the main free choices are implementation thresholds in the speech analysis.

free parameters (2)
  • utterance_duration_threshold = 125 ms
    Utterances under 125 milliseconds are excluded as non-informative in Section 3.2.2, which affects the computed speaking rate and pause metrics.
  • speech_segmentation_intervals = 60 seconds coarse, 10 seconds fine
    The two-level segmentation intervals in Section 3.2.2 are chosen by hand without a sensitivity analysis, and they influence the reported vocal dynamics.
assumptions (4)
  • domain assumption Pretrained CV models transfer to the Singapore classroom environment without fine-tuning
    Section 3.2.1 relies on pretrained YOLOv8, DeepOCSORT, ST-GCN, SAM/MobileSAM, and mmpose weights; no local fine-tuning or validation on this classroom data is reported.
  • domain assumption Low-level teacher actions and vocal metrics are valid proxies for instructional behavior
    Sections 3.2 and 3.3 construct the dashboard summaries from action labels and vocal statistics, but no ground-truth comparison against expert pedagogical coding is provided.
  • domain assumption Literature-based speech norms apply to Singapore higher education
    Section 3.3.1 benchmarks speaking rate at 140 wpm, STI at 0.75, and monotony at 1.0 using Western sources, without local validation of these norms.
  • domain assumption Feedback from eight NIE researchers is representative of broader teacher-trainer opinion
    Section 3.3.3 describes an informal evaluation with a small convenience sample from collaborating groups; the paper itself acknowledges that a standardized protocol is needed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the development of an AI performance and behavioural measures for teaching and classroom management." pith.science (2026). https://pith.science/paper/JD3E5LMP

@misc{pith2026250611143,
  author       = {Pith},
  title        = {Pith review of: On the development of an AI performance and behavioural measures for teaching and classroom management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JD3E5LMP}},
  note         = {Machine review of arXiv:2506.11143}
}
read the original abstract

This paper presents a two-year research project focused on developing AI-driven measures to analyze classroom dynamics, with particular emphasis on teacher actions captured through multimodal sensor data. We applied real-time data from classroom sensors and AI techniques to extract meaningful insights and support teacher development. Key outcomes include a curated audio-visual dataset, novel behavioral measures, and a proof-of-concept teaching review dashboard. An initial evaluation with eight researchers from the National Institute for Education (NIE) highlighted the system's clarity, usability, and its non-judgmental, automated analysis approach -- which reduces manual workloads and encourages constructive reflection. Although the current version does not assign performance ratings, it provides an objective snapshot of in-class interactions, helping teachers recognize and improve their instructional strategies. Designed and tested in an Asian educational context, this work also contributes a culturally grounded methodology to the growing field of AI-based educational analytics.

Figures

Figures reproduced from arXiv: 2506.11143 by the authors.

Figure 1
Figure 1. Camera placement in the classroom Capturing high-quality audio was equally critical, particularly since not all microphones were close-talk. Although the classroom was relatively quiet, reverberation posed challenges for distant mi￾crophone recordings. To address this, we relied on a robust micro￾phone infrastructure drawing on our previous work in acoustically 2 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Tracking system The interface consists of multiple rows representing different recording streams. As shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Annotation app 3 Output results 3.1 Audio-Visual annotated data collection Building on the data collection and annotation processes described in Sections 2.1 and 2.4.2, the resulting dataset comprises curated, time-synchronized audio-visual recordings of classroom teaching interactions in higher education contexts in Singapore. It integrates footage from multiple capture devices and includes time-aligned teacher act… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Summary screen (1) - teaching behaviors [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Summary screen (2) - spatial movement Positioned in the upper left section of the summary screen, [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Review screen (left) - video playback & position and [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 6
Figure 6. Figure 6: Summary screen (3) - speech & movement charac [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 9
Figure 9. Figure 9: Summary screen [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 8
Figure 8. Figure 8: Review screen (right) - latest teacher movement and [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 10
Figure 10. Figure 10: Monitor Screen 4 Conclusion and future work This study explored using AI to generate quantifiable insights into teacher behavior via multimodal sensor data. With our curated dataset, we identified key visual and vocal indicators of classroom practice and showed how AI…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 28 canonical work pages

  1. [1]

    ARKit Face Blendshapes (Perfect Sync)

    n.d. ARKit Face Blendshapes (Perfect Sync). Retrieved May 28, 2025 from https://arkit-face-blendshapes.com/

    work page 2025

  2. [2]

    Boersma and D

    P. Boersma and D. Weenink. 2012. Praat: Doing phonetics by computer (Version 5.1.26) [Computer program]. Retrieved May 28, 2025 from http://www.praat.org/

    work page 2012

  3. [3]

    P. Büllow. 2023. Voice Analysis – The Complete Pocket Guide. Retrieved May 28, 2025 from https://imotions.com/blog/learning/research-fundamentals/voice- analysis-the-complete-pocket-guide/

    work page 2023

  4. [4]

    Chong, K.M

    T.Y. Chong, K.M. Tan, K.K. Teh, C.H. You, H. Sun, and H.D. Tran. 2019. The I2R’s ASR System for the VOiCES from a Distance Challenge 2019. InProceedings of the 20th Annual Conference of the International Speech Communication Association, Interspeech 2019. Graz, Austria, pp. 2458–2462. doi:10.21437/Interspeech.2019- 2130

    work page doi:10.21437/interspeech.2019- 2019

  5. [5]

    MMPose Contributors. 2020. OpenMMLab Pose Estimation Toolbox and Benchmark. Retrieved May 28, 2025 from https://github.com/open-mmlab/mmpose

    work page 2020

  6. [6]

    Cosbey, A

    R. Cosbey, A. Wusterbarth, and B. Hutchinson. 2019. Deep Learning for Classroom Activity Detection from Audio. InProceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2019) . IEEE, Brighton, UK, pp.3727–3731. doi:10.1109/ICASSP.2019.8683365

    arXiv 2019

  7. [7]

    Donnelly, N

    P.J. Donnelly, N. Blanchard, B. Samei, A.M. Olney, X. Sun, B. Ward, S. Kelly, M. Nystran, and S.K. D’Mello. 2016. Automatic Teacher Modeling from Live Class- room Audio. In Proceedings of the 2016 Conference on User Modeling Adaptation and Personalization (UMAP ’16) . ACM, Halifax, Nova Scotia, Canada, pp.45–53. doi:10.1145/2930238.2930250

    work page arXiv 2016

  8. [8]

    Dragon, I

    T. Dragon, I. Arroyo, B.P. Woolf, W. Burleson, R. el Kaliouby, and H. Eydgahi

    work page

Show all 37 references
  1. [9]

    K. Du, J. Yang, K. Hu, X. Xie, J.S. He, X. Huang, and L. Wang. 2023. Classroom Behavior Recognition of Students and Teaching Quality Evaluation Method Based on YOLOv5 Model. In 2023 7th Asian Conference on Artificial Intelligence Technology (ACAIT). Quzhou, China, pp.303–309. ...

    2023

  2. [10]

    D’Haro, A.I

    L.F. D’Haro, A.I. Niculescu, C. Cai, S. Nair, R.E. Banchs, A. Knoll, and H. Li. 2017. An integrated framework for multimodal human-robot interaction. In Proc. of Asia-Pacific Signal and Information Processing Association Annual Summit and Conference. IEEE (APSIPA ASC), Kuala L...

    2017

  3. [11]

    Fan and T

    Y. Fan and T. Wang. 2024. SCB-Dataset3: A Benchmark for Detecting Student Classroom Behavior. arXiv:2310.02522 [cs.CV] https://arxiv.org/abs/2310.02522

    2024 arXiv

  4. [12]

    H. Gao, E. Bozkir, Ph. Stark, P. Goldberg, G. Meixner, E. Kasneci, and R. Göllner

  5. [13]

    Holstein, G

    K. Holstein, G. Hong, M. Tegene, B.M. McLaren, and V. Aleven. 2018. The classroom as a dashboard: co-designing wearable cognitive augmentation for K- 12 teachers. InProceedings of the 8th International Conference on Learning Analytics and Knowledge (LAK ’18) . Association for ...

    2018

  6. [14]

    Jocher, A

    G. Jocher, A. Chaurasia, and J. Qiu. 2023.Ultralytics YOLO (Version 8.0.0). Retrieved May 28, 2025 from https://github.com/ultralytics/ultralytics

    2023

  7. [15]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, and R. Girshick

  8. [16]

    Leccese, M

    F. Leccese, M. Rocca, and G. Salvadori. 2018. Fast estimation of Speech Trans- mission Index using the Reverberation Time: Comparison between predictive equations for educational rooms of different sizes. Applied Acoustics 140, 6 (2018), pp.143–149. doi:10.1016/j.apacoust.2018.05.019

    2018 doi

  9. [17]

    H. Li, Y. Kang, W. Ding, S. Yang, S. Yang, G.Y. Huang, and Z. Liu. 2020. Multimodal Learning for Classroom Activity Detection. InProceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2020) . Barcelona, Spain, pp.9234–9238. doi:10.1...

    2020

  10. [18]

    In Proceedings of the IEEE/CVF International Conference on Computer Vision

    Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision. Paris, France, pp.4015–4026

  11. [19]

    Niculescu, M.Q

    A.I. Niculescu, M.Q. Lim, S.A. Wibowo, K.H. Yeo, B.P. Lim, M. Popow, D. Chia, and R.E. Banchs. 2015. Designing IDA – an Intelligent driver assistant for smart city parking in Singapore. InProc. of the 15th IFIP Conference on Human Computer Interaction (INTERACT) (LNCS) . Sprin...

    2015

  12. [20]

    Niculescu, B

    A.I. Niculescu, B. van Dijk, A. Nijholt, D.K. Limbu, S.L. See, and A.H.Y. Wong

  13. [21]

    Maggiolino, A

    G. Maggiolino, A. Ahmad, J. Cao, and K. Kitani. 2023. Deep oc-sort: Multi- pedestrian tracking by adaptive re-identification. In 2023 IEEE International Con- ference on Image Processing (ICIP) . IEEE, Kuala Lumpur, Malaysia, pp.3025–3029

    2023

  14. [22]

    Niculescu, B

    A.I. Niculescu, B. Wadhwa, and E. Quek. 2016. Technologies for the future: evaluating a voice enabled smart city parking application. In Proc. of the 4th Int. Conference on User Science and Engineering (i-USEr) . IEEE, Melaka, Malaysia, 46–50

    2016

  15. [23]

    Niculescu, G.M

    A.I. Niculescu, G.M. White, S.L. See, R.U. Waloejo, and Y. Kawaguchi. 2008. Impact of English regional accents on user acceptance of voice user interfaces. In Proceedings of the 5th Nordic conference on Human-computer interaction: building bridges - NordiCHI ’08. ACM, Lund, Sw...

    2008

  16. [24]

    Owens, S.B

    M.T. Owens, S.B. Seidel, M. Wong, T.E. Bejines, S. Lietz, J.R. Perez, S. Sit, Z-S. Subedar, G.N. Acker, and S.F. Akana. 2017. Classroom sound can be used to classify teaching practices in college science courses. Proceedings of the National Academy of Sciences 114, 12 (2017), ...

    2017 doi

  17. [25]

    Niculescu, B

    A.I. Niculescu, B. Wadhwa, and E. Quek. 2016. Smart city technologies: design and evaluation of an intelligent driving assistant for smart parking. Int. Journal on Advanced Science, Engineering and Information Technology (IJASEIT) (2016), pp.143–149. doi:10.18517/IJASEIT.6.6.1473

    2016 doi

  18. [26]

    Rotgans and H.G

    J.I. Rotgans and H.G. Schmidt. 2011. The role of teachers in facilitating situational interest in an active-learning classroom. Teaching and Teacher Education 27, 1 (2011), pp.37–42. doi:10.1016/j.tate.2010.06.025

    2011 doi

  19. [27]

    Y. Shen, J. Zhang, and Y. Li. 2023. Behavior Recognition of Teachers and Stu- dents in the Smart Classroom Based on Deep Learning. In Proceedings of the 4th International Conference on Information Science and Education (ICISE-IE) . IEEE, Zhanjiang, China, pp.345–349. doi:10.11...

    2023

  20. [28]

    Steeneken and T

    H.J. Steeneken and T. Houtgast. 1980. A physical method for measuring speech- transmission quality. The Journal of the Acoustical Society of America 67, 1 (1980), pp.318—-326

    1980

  21. [29]

    S. Ren, K. He, R. Girshick, and J. Sun. 2017. Faster R-CNN: Towards Real-time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 39, 6 (2017), pp.1137–1149. doi:10.1109/TPAMI. 2016.2577031

    2017

  22. [30]

    Z. Wang, J. Yao, C. Zeng, W. Wu, H. Xu, and Y. Yang. 2022. YOLOv5 Enhanced Learning Behavior Recognition and Analysis in Smart Classroom with Multiple Students. InProceedings of the International Conference on Intelligent Education and Intelligent Research (IEIR). IEEE, Wuhan,...

    2022

  23. [31]

    S. Yan, Y. Xiong, and D. Lin. 2018. Spatial temporal graph convolutional networks for skeleton-based action recognition. In Proceedings of the AAAI conference on artificial intelligence, Vol. 32. New Orleans, Louisiana, U.S

    2018

  24. [32]

    Zhang, D

    C. Zhang, D. Han, Y. Qiao, J. U. Kim, S. H. Bae, S. Lee, and C. S. Hong. 2023. Faster segment anything: Towards lightweight SAM for mobile applications. arXiv preprint:2306.14289 (2023)

    2023 arXiv

  25. [33]

    Tauroza and D

    S. Tauroza and D. Allison. 1990. Speech Rates in British English. Applied Linguis- tics 11, 1 (1990), pp.90—-105

    1990

  26. [37]

    S. Zionts. 1979. MCDM: If Not a Roman Numeral, Then What? Interfaces 9, 4 (1979), pp.94—-101. 7

    1979

  27. [2008]

    In Intelligent Tutoring Systems , B

    Viewing Student Affect and Learning through Classroom Observation and Physical Sensors. In Intelligent Tutoring Systems , B. P. Woolf, E. Aïmeur, R. Nkambou, and S. Lajoie (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, pp.29–39

  28. [2010]

    Socializing with Olivia, the Youngest Robot Receptionist Outside the Lab. In Proc. of the Int. Conference in Social Robotics (LNAI, Vol. 6414) , S.S. Ge et al. (Ed.). Springer-Verlag, Berlin, Germany, pp.50–62

  29. [2023]

    In 2023 IEEE International Symposium on Mixed and Augmented Reality (ISMAR)

    Detecting Teacher Expertise in an Immersive VR Classroom: Leveraging Fused Sensor Data with Explainable Machine Learning Models. In 2023 IEEE International Symposium on Mixed and Augmented Reality (ISMAR) . Sydney, Australia, pp.683–692. doi:10.1109/ISMAR59233.2023.00083

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.