arxiv: 2604.19834 · v1 · submitted 2026-04-21 · 💻 cs.CV

Recognition: unknown

KD-Judge: A Knowledge-Driven Automated Judge Framework for Functional Fitness Movements on Edge Devices

Arun Iyengar, Fan Li, Lanyu Xu, Lucas Alves, Shaibal Saha, Yunge Li

Authors on Pith no claims yet

Pith reviewed 2026-05-10 03:31 UTC · model grok-4.3

classification 💻 cs.CV

keywords functional fitnessautomated judgingrule-based systemsedge devicespose estimationrepetition analysisknowledge-driven AILLM pipeline

0 comments

The pith

KD-Judge converts fitness rulebooks into executable code that a computer uses to judge exercise repetitions deterministically on edge devices.

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

The paper presents KD-Judge as a way to automate judging of functional fitness movements by turning unstructured rule standards into machine-readable instructions. It relies on large language models with retrieval and step-by-step reasoning to structure the rules, then applies them through pose analysis and kinematic checks to decide if each repetition is valid. This creates a transparent, rule-based alternative to opaque scoring models. Tests on the CFRep dataset show reliable rule structuring and accurate assessments that run faster than real time, with added caching that cuts computation sharply on hardware like the Jetson AGX Xavier.

Core claim

KD-Judge converts unstructured rulebook standards into executable representations using an LLM-based retrieval-augmented generation and chain-of-thought rule-structuring pipeline, incorporates the structured rules into a deterministic rule-based judging system with pose-guided kinematic reasoning to assess rep validity and temporal boundaries, and applies a dual caching strategy to achieve faster-than-real-time execution and substantial speedups on edge devices.

What carries the argument

The LLM-based retrieval-augmented generation and chain-of-thought pipeline that structures rules, combined with deterministic pose-guided kinematic reasoning and selective dual caching for edge efficiency.

If this is right

Reliable rule-structuring and accurate rep-level assessment on the CFRep dataset.
Faster-than-real-time execution with real-time factor below 1 on tested hardware.
Speedups of up to 3.36 times for pre-recorded videos and 15.91 times for live streams on resource-limited edge devices when caching is used.
Transparent and deterministic judgments that can complement human judging in training and competition settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same rule-structuring approach could apply to other rule-heavy domains such as gymnastics or weightlifting competitions where consistency matters.
Caching mechanisms for redundant pose computations might transfer to other real-time vision tasks on edge devices facing similar data streams.
Integration into fitness apps could provide immediate, rule-based feedback during workouts without needing constant cloud access.

Load-bearing premise

The LLM-based retrieval-augmented generation and chain-of-thought rule-structuring pipeline accurately converts unstructured rulebook standards into complete, error-free executable representations that match human intent.

What would settle it

A direct comparison on movements outside the CFRep dataset where the system's rep validity decisions are checked against a panel of human judges using the identical rulebooks, looking for systematic mismatches in boundary detection or validity calls.

Figures

Figures reproduced from arXiv: 2604.19834 by Arun Iyengar, Fan Li, Lanyu Xu, Lucas Alves, Shaibal Saha, Yunge Li.

**Figure 3.** Figure 3: Overview of Stage 1 RAG pipeline for converting [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Figure 4: Overview of Stage 3 Judge System: The judge system [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Overview of dual strategy cache mechanism. Frame [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Cosine similarity threshold comparison for IF3 and [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 7.** Figure 7: Rep-level latency and speedup (× relative to the w/o cache baseline) across different pose estimation models using cache-based techniques on edge devices. W/o Cache DC RTC Combined RTF=1 HRNet-W32 RTMPose-L RTMPose-X RTMO-L 0.0 1.0 1.2 RTF (a) RTX 3080 HRNet-W32 RTMPose-L RTMPose-X RTMO-L 0 1 2 RTF (b) AGX Xavier [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: RTF analysis across pose models with caching strate [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗

**Figure 9.** Figure 9: Adaptability analysis of KD-Judge on double under. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗

read the original abstract

Functional fitness movements are widely used in training, competition, and health-oriented exercise programs, yet consistently enforcing repetition (rep) standards remains challenging due to subjective human judgment, time constraints, and evolving rules. Existing AI-based approaches mainly rely on learned scoring or reference-based comparisons and lack explicit rule-based, limiting transparency and deterministic rep-level validation. To address these limitations, we propose KD-Judge, a novel knowledge-driven automated judging framework for functional fitness movements. It converts unstructured rulebook standards into executable, machine-readable representations using an LLM-based retrieval-augmented generation and chain-of-thought rule-structuring pipeline. The structured rules are then incorporated by a deterministic rule-based judging system with pose-guided kinematic reasoning to assess rep validity and temporal boundaries. To improve efficiency on edge devices, including a high-performance desktop and the resource-constrained Jetson AGX Xavier, we introduce a dual strategy caching mechanism that can be selectively applied to reduce redundant and unnecessary computation. Experiments demonstrate reliable rule-structuring performance and accurate rep-level assessment, with judgment evaluation conducted on the CFRep dataset, achieving faster-than-real-time execution (real-time factor (RTF) < 1). When the proposed caching strategy is enabled, the system achieves up to 3.36x and 15.91x speedups on resource-constrained edge device compared to the non-caching baseline for pre-recorded and live-streaming scenarios, respectively. These results show that KD-Judge enables transparent, efficient, and scalable rule-grounded rep-level analysis that can complement human judging in practice.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

KD-Judge pairs LLM rule structuring with deterministic pose checks and edge caching, but the accuracy of the rule translation step is asserted without supporting numbers.

read the letter

The paper's main contribution is a pipeline that feeds fitness rulebooks into an LLM with RAG and chain-of-thought to produce executable constraints, then runs deterministic kinematic checks on pose data to score reps, plus selective caching to hit real-time factors below 1 on a Jetson AGX Xavier. The caching delivers reported speedups of 3.36x for pre-recorded video and 15.91x for live streams versus the uncached version. That combination is new for this domain and moves past pure learned scoring or simple reference matching mentioned in the abstract. The engineering description of the dual-cache strategy and the hardware measurements are concrete and useful for anyone targeting constrained devices. The work is aimed at applied sports-tech or health-monitoring groups that want transparent, rule-grounded judgments rather than black-box models. Readers who need a working system for functional fitness on edge hardware could pull practical ideas from the caching and pipeline layout. The soft spot is the missing quantitative check on the LLM step. The abstract calls the rule structuring reliable and the rep assessment accurate on the CFRep dataset, yet supplies no precision figures, human-expert agreement rates, or error cases for how well the generated rules match the original standards. Without those, the downstream kinematic results rest on an unverified translation; one consistent mis-mapping of a timing or angle constraint would undermine the claims. If the full manuscript adds condition-level metrics or failure analysis, the soundness improves; as presented in the abstract alone, the central accuracy assertion stays unanchored. I would bring this to a reading group for the caching and rule-pipeline details, but I would not cite it in my own work until the validation numbers appear. It deserves peer review because the applied problem is real and the hardware results are measurable, even if the rule-fidelity part needs tightening.

Referee Report

2 major / 0 minor

Summary. The paper introduces KD-Judge, a knowledge-driven framework for automated judging of functional fitness movements. It uses an LLM-based retrieval-augmented generation and chain-of-thought pipeline to convert unstructured rulebook standards into executable machine-readable representations. These structured rules feed a deterministic rule-based judging system that performs pose-guided kinematic reasoning to determine repetition validity and temporal boundaries. A dual caching strategy is proposed to reduce redundant computation on edge devices. Evaluation on the CFRep dataset reports faster-than-real-time execution (RTF < 1) and speedups of up to 3.36x (pre-recorded) and 15.91x (live-streaming) on a Jetson AGX Xavier compared to a non-caching baseline.

Significance. If the rule-structuring step is shown to faithfully capture human intent, the framework would provide a transparent, deterministic alternative to learned or reference-based scoring methods for fitness assessment. The caching mechanism directly addresses deployment constraints on resource-limited hardware, potentially enabling practical use in training, competition, or health programs where real-time rule enforcement is needed.

major comments (2)

[Abstract] Abstract: The claims of 'reliable rule-structuring performance' and 'accurate rep-level assessment' are presented without any quantitative metrics (condition-level precision, human-expert agreement, failure-case analysis, or baseline comparisons). This is load-bearing for the central claim because the deterministic kinematic judgments rest entirely on the fidelity of the LLM+RAG+CoT translation step; a systematic mis-mapping of temporal or kinematic constraints would invalidate all downstream results.
[Experiments] Experiments section: No dataset details (number of movements, rules, ground-truth annotations, or split statistics) or error analysis for the CFRep evaluation are supplied, preventing assessment of whether the reported speedups and RTF < 1 actually support the accuracy claims under realistic rule variations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. These observations highlight important aspects of clarity and substantiation that will strengthen the paper. We provide point-by-point responses below and will revise the manuscript to address the concerns.

read point-by-point responses

Referee: [Abstract] Abstract: The claims of 'reliable rule-structuring performance' and 'accurate rep-level assessment' are presented without any quantitative metrics (condition-level precision, human-expert agreement, failure-case analysis, or baseline comparisons). This is load-bearing for the central claim because the deterministic kinematic judgments rest entirely on the fidelity of the LLM+RAG+CoT translation step; a systematic mis-mapping of temporal or kinematic constraints would invalidate all downstream results.

Authors: We acknowledge that the abstract asserts reliable rule-structuring and accurate assessment without accompanying quantitative metrics, and that this is a critical point given the reliance on the LLM+RAG+CoT pipeline for downstream determinism. While the CFRep dataset evaluation provides end-to-end evidence of rep-level performance, we agree that direct metrics for the structuring step are necessary to fully support the claims. In the revised manuscript, we will add quantitative results including condition-level precision/recall for rule extraction, human-expert agreement scores on structured rules, failure-case analysis of mis-mapped constraints, and comparisons to baseline rule-structuring approaches. These will be incorporated into the Experiments section, with the abstract updated to reference the new metrics. revision: yes
Referee: [Experiments] Experiments section: No dataset details (number of movements, rules, ground-truth annotations, or split statistics) or error analysis for the CFRep evaluation are supplied, preventing assessment of whether the reported speedups and RTF < 1 actually support the accuracy claims under realistic rule variations.

Authors: We agree that the Experiments section lacks sufficient dataset specifics and error analysis, which limits evaluation of how the accuracy claims hold under rule variations alongside the efficiency results. The current text references the CFRep dataset for judgment evaluation but omits details on the number of movements and rules, ground-truth annotation process, split statistics, and any error breakdown. In the revision, we will expand the section to include these elements: dataset composition (movements, rules, and variations), annotation methodology and inter-annotator reliability, data splits, and error analysis (e.g., accuracy stratified by movement type or rule complexity). This will enable readers to assess whether the RTF < 1 and speedups (3.36x pre-recorded, 15.91x live) are supported by robust accuracy under realistic conditions. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper describes an LLM RAG+CoT pipeline that converts external rulebooks into executable rules, followed by a deterministic pose-based kinematic judge and a caching optimization, all evaluated on the independent CFRep dataset with measured RTF and speedup numbers on Jetson hardware. No equations, parameters, or claims reduce by construction to their own inputs; there are no self-citations that bear the central load, no fitted quantities renamed as predictions, and no ansatz or uniqueness results imported from prior author work. The performance assertions are grounded in external data and timing measurements rather than tautological re-derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework depends on standard assumptions about pose estimation reliability and LLM rule extraction fidelity but introduces no explicit free parameters, new physical entities, or ad-hoc axioms beyond domain-standard computer vision and language model capabilities.

axioms (2)

domain assumption Pose estimation from video provides sufficiently accurate kinematic data for validating fitness movement rules.
The judging system relies on pose-guided kinematic reasoning to determine rep validity and temporal boundaries.
domain assumption LLM retrieval-augmented generation with chain-of-thought produces faithful machine-readable rule representations from unstructured text.
Central pipeline step for converting rulebooks into executable form.

pith-pipeline@v0.9.0 · 5597 in / 1437 out tokens · 67578 ms · 2026-05-10T03:31:38.741798+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 11 canonical work pages · 5 internal anchors

[1]

Machine learning methods in sport injury prediction and prevention: a systematic review,

H. Van Eetvelde, L. D. Mendonça, C. Ley, R. Seil, and T. Tischer, “Machine learning methods in sport injury prediction and prevention: a systematic review,” Journal of experimental orthopaedics, vol. 8, no. 1, p. 27, 2021

2021
[2]

Machine learning in human movement biomechanics: Best practices, common pitfalls, and new opportunities,

E. Halilaj, A. Rajagopal, M. Fiterau, J. L. Hicks, T. J. Hastie, and S. L. Delp, “Machine learning in human movement biomechanics: Best practices, common pitfalls, and new opportunities,” Journal of biomechanics, vol. 81, pp. 1–11, 2018

2018
[3]

A survey of human gait- based artificial intelligence applications,

E. J. Harris, I.-H. Khoo, and E. Demircan, “A survey of human gait- based artificial intelligence applications,” Frontiers in Robotics and AI, vol. 8, p. 749274, 2022

2022
[4]

Deep gait recognition: A sur- vey,

A. Sepas-Moghaddam and A. Etemad, “Deep gait recognition: A sur- vey,” IEEE transactions on pattern analysis and machine intelligence, vol. 45, no. 1, pp. 264–284, 2022

2022
[5]

Human action recognition: A taxonomy-based survey, updates, and opportunities,

M. G. Morshed, T. Sultana, A. Alam, and Y .-K. Lee, “Human action recognition: A taxonomy-based survey, updates, and opportunities,” Sensors, vol. 23, no. 4, p. 2182, 2023

2023
[6]

Finediving: A fine-grained dataset for procedure-aware action quality assessment,

J. Xu, Y . Rao, X. Yu, G. Chen, J. Zhou, and J. Lu, “Finediving: A fine-grained dataset for procedure-aware action quality assessment,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 2949–2958

2022
[7]

Finegym: A hierarchical video dataset for fine-grained action understanding,

D. Shao, Y . Zhao, B. Dai, and D. Lin, “Finegym: A hierarchical video dataset for fine-grained action understanding,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 2616–2625

2020
[8]

Aifit: Automatic 3d human-interpretable feedback models for fitness training,

M. Fieraru, M. Zanfir, S. C. Pirlea, V . Olaru, and C. Sminchisescu, “Aifit: Automatic 3d human-interpretable feedback models for fitness training,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 9919–9928

2021
[9]

Retrieval- augmented generation for knowledge-intensive nlp tasks,

P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. Küttler, M. Lewis, W.-t. Yih, T. Rocktäschel et al., “Retrieval- augmented generation for knowledge-intensive nlp tasks,” Advances in neural information processing systems, vol. 33, pp. 9459–9474, 2020

2020
[10]

Chain-of-thought prompting elicits reasoning in large language models,

J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhou et al., “Chain-of-thought prompting elicits reasoning in large language models,” Advances in neural information processing systems, vol. 35, pp. 24 824–24 837, 2022

2022
[11]

Repval: A skeleton-based validation system for functional fitness repetition on edge devices,

L. Alves, F. Li, and L. Xu, “Repval: A skeleton-based validation system for functional fitness repetition on edge devices,” in Proceedings of the Tenth ACM/IEEE Symposium on Edge Computing, ser. SEC ’25. New York, NY , USA: Association for Computing Machinery, 2025. [Online]. Available: https://doi.org/10.1145/3769102.3774242

work page doi:10.1145/3769102.3774242 2025
[12]

Multi- modal large language model with rag strategies in soccer commentary generation,

X. Li, Y . He, S. Zu, Z. Li, T. Shi, Y . Xie, and K. Zhang, “Multi- modal large language model with rag strategies in soccer commentary generation,” in 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV), 2025, pp. 6197–6206

2025
[13]

Court to conversation: Tactical badminton analysis via computer vision and rag-enhanced llms,

K. Bharadwaj and G. Srinivasa, “Court to conversation: Tactical badminton analysis via computer vision and rag-enhanced llms,” Knowledge-Based Systems, vol. 333, p. 115027, 2026. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S0950705125020659

2026
[14]

Sportsgpt: An llm-driven framework for interpretable sports motion assessment and training guidance.arXiv preprint arXiv:2512.14121, 2025

W. Tian, R. Lin, H. Zheng, Y . Yang, G. Wu, Z. Zhang, and Z. Zhang, “Sportsgpt: An llm-driven framework for interpretable sports motion assessment and training guidance,” arXiv preprint arXiv:2512.14121, 2025

work page arXiv 2025
[15]

Rag-har: Retrieval augmented generation-based human activity recognition,

N. Sivaroopan, H. Karunarathna, C. Madarasingha, A. Jayasumana, and K. Thilakarathna, “Rag-har: Retrieval augmented generation-based human activity recognition,” arXiv preprint arXiv:2512.08984, 2025

work page internal anchor Pith review arXiv 2025
[16]

Extracting accurate materials data from research papers with conversational language models and prompt engi- neering,

M. P. Polak and D. Morgan, “Extracting accurate materials data from research papers with conversational language models and prompt engi- neering,” Nature Communications, vol. 15, no. 1, p. 1569, 2024

2024
[17]

Enhancing structured data generation with gpt-4o evaluating prompt efficiency across prompt styles,

A. Elnashar, J. White, and D. C. Schmidt, “Enhancing structured data generation with gpt-4o evaluating prompt efficiency across prompt styles,” Frontiers in Artificial Intelligence, vol. 8, p. 1558938, 2025

2025
[18]

A prompt engineering approach for structured data extraction from unstructured text using conversational llms,

A. Vijayan, “A prompt engineering approach for structured data extraction from unstructured text using conversational llms,” in Proceedings of the 2023 6th International Conference on Algorithms, Computing and Artificial Intelligence, ser. ACAI ’23. New York, NY , USA: Association for Computing Machinery, 2024, p. 183–189. [Online]. Available: https://doi...

work page doi:10.1145/3639631.3639663 2023
[19]

Checkmanual: A new challenge and benchmark for manual-based appliance manipula- tion,

Y . Long, J. Zhang, M. Pan, T. Wu, T. Kim, and H. Dong, “Checkmanual: A new challenge and benchmark for manual-based appliance manipula- tion,” in Proceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 22 595–22 604

2025
[20]

Learning sparse temporal video mapping for action quality assessment in floor gymnastics,

S. Zahan, G. M. Hassan, and A. Mian, “Learning sparse temporal video mapping for action quality assessment in floor gymnastics,” IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1–11, 2024

2024
[21]

Lucidaction: A hierarchical and multi-model dataset for comprehensive action quality assessment,

L. Dong, W. Wang, Y . Qiao, and X. Sun, “Lucidaction: A hierarchical and multi-model dataset for comprehensive action quality assessment,” Advances in Neural Information Processing Systems, vol. 37, pp. 96 468–96 482, 2024

2024
[22]

Ai coach: Deep human pose estimation and analysis for personalized athletic training assistance,

J. Wang, K. Qiu, H. Peng, J. Fu, and J. Zhu, “Ai coach: Deep human pose estimation and analysis for personalized athletic training assistance,” in Proceedings of the 27th ACM international conference on multimedia, 2019, pp. 374–382

2019
[23]

Linq-Embed-Mistral technical report.arXiv:2412.03223, 2024

C. Choi, J. Kim, S. Lee, J. Kwon, S. Gu, Y . Kim, M. Cho, and J.-y. Sohn, “Linq-embed-mistral technical report,”arXiv preprint arXiv:2412.03223, 2024

work page arXiv 2024
[24]

Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration,

J. Lin, J. Tang, H. Tang, S. Yang, W.-M. Chen, W.-C. Wang, G. Xiao, X. Dang, C. Gan, and S. Han, “Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration,” Proceedings of machine learning and systems, vol. 6, pp. 87–100, 2024

2024
[25]

Efficient memory management for large language model serving with pagedattention,

W. Kwon, Z. Li, S. Zhuang, Y . Sheng, L. Zheng, C. H. Yu, J. Gonzalez, H. Zhang, and I. Stoica, “Efficient memory management for large language model serving with pagedattention,” in Proceedings of the 29th symposium on operating systems principles, 2023, pp. 611–626

2023
[26]

FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning

T. Dao, “Flashattention-2: Faster attention with better parallelism and work partitioning,” arXiv preprint arXiv:2307.08691, 2023

work page internal anchor Pith review arXiv 2023
[27]

GPT-4 Technical Report

J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[28]

Openmmlab pose estimation toolbox and benchmark,

M. Contributors, “Openmmlab pose estimation toolbox and benchmark,” https://github.com/open-mmlab/mmpose, 2020

2020
[29]

Human pose- based estimation, tracking and action recognition with deep learning: A survey,

L. Zhou, X. Meng, Z. Liu, M. Wu, Z. Gao, and P. Wang, “Human pose- based estimation, tracking and action recognition with deep learning: A survey,” arXiv preprint arXiv:2310.13039, 2023

work page arXiv 2023
[30]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Bi et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” arXiv preprint arXiv:2501.12948, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[31]

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

G. Comanici, E. Bieber, M. Schaekermann, I. Pasupat, N. Sachdeva, I. Dhillon, M. Blistein, O. Ram, D. Zhang, E. Rosen et al., “Gem- ini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities,” arXiv preprint arXiv:2507.06261, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[32]

Rtmdet: An empirical study of designing real-time object detectors.arXiv preprint arXiv:2212.07784,

C. Lyu, W. Zhang, H. Huang, Y . Zhou, Y . Wang, Y . Liu, S. Zhang, and K. Chen, “Rtmdet: An empirical study of designing real-time object detectors,” arXiv preprint arXiv:2212.07784, 2022

work page arXiv 2022
[33]

Intraclass correlations: uses in assessing rater reliability

P. E. Shrout and J. L. Fleiss, “Intraclass correlations: uses in assessing rater reliability.” Psychological bulletin, vol. 86, no. 2, p. 420, 1979

1979
[34]

Activitynet: A large-scale video benchmark for human activity under- standing,

F. Caba Heilbron, V . Escorcia, B. Ghanem, and J. Carlos Niebles, “Activitynet: A large-scale video benchmark for human activity under- standing,” in Proceedings of the ieee conference on computer vision and pattern recognition, 2015, pp. 961–970

2015
[35]

A comparative study of visual and auditory reaction times on the basis of gender and physical activity levels of medical first year students,

A. Jain, R. Bansal, A. Kumar, and K. Singh, “A comparative study of visual and auditory reaction times on the basis of gender and physical activity levels of medical first year students,” International journal of applied and basic medical research, vol. 5, no. 2, pp. 124–127, 2015

2015