Pith. sign in

REVIEW 4 major objections 4 minor 6 cited by

DVD: A Comprehensive Dataset for Advancing Violence Detection in Real-World Scenarios

T0 review · 4 major / 4 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 This paper introduces DVD, a 500-video, 2.7M-frame benchmark with frame-level violence labels and detailed metadata, built to overcome the coarse, small, and homogeneous datasets that currently dominate violence detection.

desk verdict A promising frame-level violence detection dataset that can't yet be verified: the reported frame count doesn't add up, no annotation-agreement stats, and no experiments. read the letter →

arxiv 2506.05372 v1 pith:KXHDSKVO submitted 2025-05-29 cs.CV

Dimitrios Kollias , Damith C. Senadeera , Jianian Zheng , Kaushal K. K. Yadav , Greg Slabaugh , Muhammad Awais , Xiaoyun Yang This is my paper · ORCID
classification cs.CV
keywords violencedetectionframe-levelannotationvideodatasetreal-worldsurveillancemetadatabenchmarkmultilingualsearchwomeninclusion
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

Existing violence-detection datasets are mostly small, clip-level, and homogeneous, so models trained on them may not transfer to real surveillance conditions. DVD is the paper's attempt to supply the missing resource: 500 long videos, about 2.7M frames, collected from public online footage in six languages, with every frame labeled violent or non-violent and detailed metadata about the event, participants, camera type, location, and sound. The videos run from 15 to 1,200 seconds, mix surveillance, bodycam, dashcam, phone, and other footage, and include women in 200 videos, addressing both realism and the gender imbalance common in earlier datasets. A sympathetic reader would care because frame-level labels and metadata allow temporal localization and situational evaluation that clip-level annotations cannot support. The database is intended for public release upon acceptance, and the paper itself is a resource proposal rather than a report of benchmark numbers.

What carries the argument

The central object is the DVD database itself: 500 videos of 15–1,200 seconds, about 2.7M frames, with a binary violent/non-violent label per frame and per-frame metadata fields for participants, event type, scene description, footage type, indoor/outdoor location, and sound association. The mechanism meant to make the database reliable is the annotation pipeline: annotators watch each complete video first, label violent segments with a small context margin, treat pauses in an ongoing confrontation as positive but extended de-escalation breaks as negative, review annotations a second time, and keep only frames on which at least three of four trained experts agree; an independent reviewer then checks labels and metadata. This pipeline is what the claim of high-quality frame-level ground truth rests on.

What would settle it

Independently re-annotate a random sample of DVD frames with fresh annotators and compare with the released labels, and audit a random sample of metadata fields against the video content. If the binary labels fail an agreement coefficient such as kappa near 0.8, or if even a small fraction of metadata descriptions contradict what is visible in the footage, the claim that DVD is a reliable frame-level ground truth would be falsified.

Watch

Extended reading notes

Core claim

On its own terms, DVD is a claim that violence detection needs a large, frame-level, richly annotated collection rather than more architectures on saturated clip sets. The dataset is built by multilingual keyword search across public online video sources, manual filtering of staged or unclear content, and a four-annotator frame-labeling protocol in which every retained label had the agreement of at least three experts, followed by an independent review. The novelties the paper lists are variable-length long videos, frame-level granularity, inclusion of women in 200 videos, and per-frame metadata covering the number of people involved, event type, scene description, footage type, indoor or outdoor setting, and sound association. The paper asserts that models trained on DVD generalize well, and it illustrates this with predictions from a network trained on DVD and on older datasets; it presents no numeric benchmark results.

Load-bearing premise

The load-bearing premise is that the frame-level violence labels and metadata are accurate and consistent; the paper describes a four-annotator protocol with three-expert agreement but reports no inter-annotator agreement statistic, and the dataset is not yet available for direct inspection.

Editorial extensions

If this is right

  • Frame-level labels turn violence detection from a clip-classification task into a temporal-localization task, so models can be scored on when and how long violence occurs, not just whether a clip is violent.
  • The per-frame metadata makes it possible to evaluate models separately for camera type, indoor versus outdoor scenes, sound association, and number of participants, which existing clip-level datasets do not support.
  • With most frames non-violent, training on DVD forces methods to handle realistic class imbalance; the paper's suggested weighted-loss training is a direct corollary.
  • Including women participants in 200 videos offers a way to test whether a violence detector generalizes beyond the male-dominated incidents that most older datasets emphasize.
  • The 55/15/30 train-validation-test split, with leakage manually checked, gives the community a ready-made evaluation protocol.

Reading between the lines

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

  • Because the metadata distinguishes camera type and lighting, DVD could also serve as a domain-shift benchmark: a model trained on bodycam footage and tested on dashcam or phone footage would directly probe the generalizability the paper claims, an experiment the paper does not run.
  • A sharper falsifier than overall accuracy would be a hard-negative audit: measuring how often the model mislabels handshakes, high-fives, hugs, or crowds walking as violent, since the paper lists these as deliberately included challenging contexts but does not quantify their prevalence.
  • The promised public release will determine whether the benchmark claim holds; until the data and inter-annotator agreement are available, the contribution is a proposal rather than a verified resource.
  • If future users train on DVD and test on clip-level datasets, DVD's long mixed videos may yield a more realistic measure of real-world performance than the near-saturated existing benchmarks.
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

4 major / 4 minor

Summary. The paper introduces DVD, a proposed video violence detection database containing 500 videos and claimed 2.7 million frame-level annotations, with rich per-frame metadata, variable video lengths and resolutions, diverse footage types, and inclusion of women. The authors argue that existing databases are limited by coarse video-level labels, small scale, lack of diversity, and missing metadata. The paper describes collection, annotation, post-processing, database properties, and a train/validation/test split, and it claims that models trained on DVD generalize well across databases. The dataset is not yet publicly released, and the paper reports no quantitative evaluation of the database itself.

Significance. If the claims are verified, DVD would fill a real gap in the violence-detection literature: it is one of the few proposed databases with frame-level annotations, long and variable-length videos, and structured metadata such as scene type, camera type, sound association, and participant counts. The detailed annotation protocol, manual leakage-prevention step, and multilingual search are constructive design choices. However, the contribution is currently unverifiable because the dataset is unreleased, the headline frame count is internally inconsistent, no inter-annotator agreement statistic is provided, and the generalization claim in Section 3.5 is unsupported by experiments. The qualitative examples and descriptive metadata are promising, but the central quantitative and quality claims need correction and evidence before the paper can be accepted.

major comments (4)
  1. [Abstract, Sec. 1, Table 1, Sec. 3.3] The headline scale claim of 2.7M frames is inconsistent with Table 1: 900K violent + 1.6M non-violent frames sum to 2.5M, not 2.7M. The 200K-frame gap is never explained. Since Section 3.2 states that only annotations on which at least three of four annotators agreed were kept, the missing frames may correspond to discarded annotations, but the paper neither says so nor reports the number. The abstract and Table 1 must be reconciled, and the composition of the 2.7M figure clarified.
  2. [Sec. 3.5] The claim that 'models trained on DVD generalize well across the same database or across other databases' is made without any reported experiment. Figure 4 shows only qualitative prediction samples from CLIP-VDNet; no evaluation protocol, metric, baseline table, or cross-dataset result is provided. The authors should either supply the quantitative results or explicitly restrict the claim to the qualitative illustration.
  3. [Sec. 3.2] No inter-annotator agreement statistic (e.g., Fleiss' kappa or pairwise agreement) is reported for the binary frame labels or for the metadata, despite the assertion that a high degree of inter-annotator agreement was achieved. Because DVD is unreleased at submission, the reader cannot inspect the labels. Reporting agreement percentages per annotation type is necessary to support the ground-truth quality claim.
  4. [Sec. 3.3, Table 2] The metadata counts in Table 2 are difficult to interpret: 1,457 outdoor scenes plus 543 indoor scenes sum to 2,000 scenes, but the database is described as 500 videos. The paper does not define the unit of a 'scene' or state whether the scene-type and sound-association counts are per frame, per segment, or per video. This ambiguity affects the claimed richness of the metadata and should be clarified with a schema and example annotations.
minor comments (4)
  1. [Sec. 3.2] The sentence ending in the bracketed citation range '[1–7, 12, 14–18, 18–36, 36–41, 41–52, 52–56, 59, 59, 60, 65, 66, 70]' appears to be an unintended reference-block artifact; it should be replaced with specific relevant citations or removed.
  2. [References] References [4] and [5] are duplicates, and references [9] and [58] describe the same Hockey Fight dataset; these should be consolidated.
  3. [Secs. 3.1, 3.4, 3.6] Sections 3.1 and 3.6 both describe keyword and multilingual collection, and Section 3.4 with Figure 3 largely repeats content from Section 3.3 with Figure 1; the presentation would be cleaner if these were consolidated.
  4. [Abstract and Sec. 3.3] The abstract states '2.7M frames' while Section 3.3 says 'around 2.7 million frames'; please state the exact frame count and how it relates to the violent/non-violent split.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper makes empirical/resource claims about a new dataset, not a derivation that reduces to its inputs.

full rationale

This is a dataset presentation paper. Its central claims are empirical and resource-based: 500 videos, 2.7M frames, frame-level annotations, metadata, and diversity properties. There is no derivation, fitted parameter, or equation that connects a claimed prediction back to its inputs, and no load-bearing use of prior work by the same authors. The reference list is unusually heavy with self-citations, including an apparent accidental citation block in Section 3.2, but none of those citations carries a load-bearing inference: no uniqueness theorem, ansatz, or fitted quantity is imported from them. The sentence in Section 3.5 that 'models trained on DVD generalize well across the same database or across other databases' is an unsupported empirical assertion, not a circular reduction, and Figure 4 is qualitative. The internal inconsistency between Table 1 (900K violent + 1.6M non-violent frames = 2.5M) and the abstract's 2.7M claim is a data consistency or verifiability problem, not circularity. Since the dataset is unreleased and no inter-annotator agreement statistic is given, external verification is currently impossible, but impossibility of verification is distinct from the paper's claims reducing to themselves. No specific circular step can be quoted and exhibited, so the circularity score is 0.

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

The central claims (dataset quality, representativeness, and annotation reliability) rest on unverifiable assumptions because the dataset and annotation files are not released and no inter-annotator metrics are reported.

assumptions (4)
  • ad hoc to paper Frame-level annotation quality: at least three of four annotators agreeing defines ground truth.
    Section 3.2 states annotations on which at least three experts agreed were kept; this assumes majority agreement equals correctness, but no agreement statistics are provided.
  • domain assumption YouTube keyword search yields representative real-world violence.
    Sections 3.1 and 3.6 describe manual filtering of staged or unclear videos; this assumes the remaining videos are genuine and representative of real-world violent events.
  • domain assumption The binary violent/non-violent distinction is well-defined across contexts and cultures.
    Annotation guidelines define violence, but interpretation is subjective; the paper filters ambiguous cases, which may hide disagreement and make the labels less transferable.
  • ad hoc to paper Manual train/validation/test partitioning prevents data leakage.
    Section 3.3 states partitioning was manually verified for leakage, but the verification procedure is not described or checkable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DVD: A Comprehensive Dataset for Advancing Violence Detection in Real-World Scenarios." pith.science (2026). https://pith.science/paper/KXHDSKVO

@misc{pith2026250605372,
  author       = {Pith},
  title        = {Pith review of: DVD: A Comprehensive Dataset for Advancing Violence Detection in Real-World Scenarios},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KXHDSKVO}},
  note         = {Machine review of arXiv:2506.05372}
}
read the original abstract

Violence Detection (VD) has become an increasingly vital area of research. Existing automated VD efforts are hindered by the limited availability of diverse, well-annotated databases. Existing databases suffer from coarse video-level annotations, limited scale and diversity, and lack of metadata, restricting the generalization of models. To address these challenges, we introduce DVD, a large-scale (500 videos, 2.7M frames), frame-level annotated VD database with diverse environments, varying lighting conditions, multiple camera sources, complex social interactions, and rich metadata. DVD is designed to capture the complexities of real-world violent events.

Figures

Figures reproduced from arXiv: 2506.05372 by the authors.

Figure 1
Figure 1. Example videos from our DVD database, showcasing diverse scenes, environments, participants, and camera types. Each video [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Distribution of video footage types in the database, with [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Example videos from the Diverse video Violence Database (DVD). The database consists of videos from a range of scenes, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: DVD video samples showcasing its labels and the predictions of CLIP-VDNet trained on DVD, RWF-2000, RLVS, and Vio Peru. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Child-Oriented AIGC Video Risk Reviewing: A Benchmark and Knowledge-Supported Iterative Reasoning Framework

    cs.CV 2026-07 reject novelty 6.0 of 10

    A multi-agent iterative-questioning framework plus a 605-video benchmark for detecting developmentally inappropriate risks in AI-generated children's videos.

  2. FuseMamba-VD: Dual Branch VideoMamba with Gated Class Token Fusion for Violence Detection

    cs.CV 2025-05 conditional novelty 6.0 of 10

    A dual-branch VideoMamba with gated class-token fusion achieves 95.85% accuracy on a newly combined violence-detection benchmark and 74.13% on DVD, with about half the parameters and FLOPs of the CUE-Net baseline.

  3. EgoSafe: A First-Person Mobile-Captured Benchmark for Visual Safety Understanding

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A new egocentric safety benchmark shows current video-language models can describe scenes well but fail at multi-step causal reasoning about blind spots and covert actions.

  4. Distance-aware Soft Prompt Guidance for Multimodal Valence-Arousal Estimation

    cs.CV 2026-03 reject novelty 5.0 of 10

    Distance-aware soft prompts over a 3×3 emotion grid with CLIP text prototypes and audio-visual GRU fusion achieve CCC_mean 0.5361 on Aff-Wild2, beating only the paper's self-defined baselines.

  5. Team RAS in 9th ABAW Competition: Multimodal Compound Expression Recognition Approach

    cs.CV 2025-07 conditional novelty 5.0 of 10

    A six-modality zero-shot pipeline with CLIP, Qwen-VL, WavLM, Mamba, and new fusion/aggregation modules reports F1 scores of 46.95 (AffWild2), 49.02 (AFEW), and 34.85 (C-EXPR-DB) without target-domain fine-tuning.

  6. TAGF: Time-aware Gated Fusion for Multimodal Valence-Arousal Estimation

    cs.MM 2025-07 reject novelty 4.0 of 10

    TAGF adds a BiLSTM-based gate that reweights recursive cross-attention outputs for valence-arousal prediction, with results slightly below several existing methods on Aff-Wild2.

Reference graph

Works this paper leans on

68 extracted references · 54 canonical work pages · cited by 6 Pith papers

  1. [1]

    Nefeli: A deep-learning detection and tracking pipeline for enhancing autonomy in advanced air mobility

    Anastasios Arsenos, Evangelos Petrongonas, Orfeas Filip- popoulos, Christos Skliros, Dimitrios Kollias, and Stefanos Kollias. Nefeli: A deep-learning detection and tracking pipeline for enhancing autonomy in advanced air mobility. Available at SSRN 4674579. 4

    work page

  2. [2]

    A large imaging database and novel deep neural ar- chitecture for covid-19 diagnosis

    Anastasios Arsenos, Dimitrios Kollias, and Stefanos Kol- lias. A large imaging database and novel deep neural ar- chitecture for covid-19 diagnosis. In2022 IEEE 14th Im- age, Video, and Multidimensional Signal Processing Work- shop (IVMSP), pages 1–5. IEEE, 2022

    work page 2022

  3. [3]

    Data-driven covid-19 detection through medical imaging

    Anastasios Arsenos, Andjoli Davidhi, Dimitrios Kollias, Panos Prassopoulos, and Stefanos Kollias. Data-driven covid-19 detection through medical imaging. In2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), pages 1–5. IEEE, 2023

    work page 2023

  4. [5]

    Common Corruptions for Enhancing and Evaluating Robustness in Air-to-Air Visual Object Detection

    Anastasios Arsenos, Vasileios Karampinis, Evangelos Petrongonas, Christos Skliros, Dimitrios Kollias, Stefanos Kollias, and Athanasios V oulodimos. Common corruptions for enhancing and evaluating robustness in air-to-air visual object detection.arXiv preprint arXiv:2405.06765, 2024

    work page Pith review arXiv 2024

  5. [6]

    Common corruptions for evaluating and enhancing robustness in air-to-air visual object detection.IEEE Robotics and Automation Letters, 2024

    Anastasios Arsenos, Vasileios Karampinis, Evangelos Petrongonas, Christos Skliros, Dimitrios Kollias, Stefanos Kollias, and Athanasios V oulodimos. Common corruptions for evaluating and enhancing robustness in air-to-air visual object detection.IEEE Robotics and Automation Letters, 2024

    work page 2024

  6. [7]

    Uncertainty- guided contrastive learning for single source domain gener- alisation

    Anastasios Arsenos, Dimitrios Kollias, Evangelos Petrong- onas, Christos Skliros, and Stefanos Kollias. Uncertainty- guided contrastive learning for single source domain gener- alisation. InICASSP 2024-2024 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), pages 6935–6939. IEEE, 2024. 4

    work page 2024

  7. [8]

    Efficient human violence recognition for surveillance in real time.Sensors, 24(2), 2024

    Herwin Alayn Huillcen Baca, Flor de Luz Palomino Val- divia, and Juan Carlos Gutierrez Caceres. Efficient human violence recognition for surveillance in real time.Sensors, 24(2), 2024. 1, 2, 3

    work page 2024

  8. [9]

    Violence detection in video using computer vision techniques

    Enrique Bermejo Nievas, Oscar Deniz Suarez, Gloria Bueno Garc´ıa, and Rahul Sukthankar. Violence detection in video using computer vision techniques. InComputer Anal- ysis of Images and Patterns: 14th International Conference, CAIP 2011, Seville, Spain, August 29-31, 2011, Proceedings, Part II 14, pages 332–339. Springer, 2011. 2, 3

    work page 2011

Show all 68 references
  1. [10]

    Cue-net: Violence detection video analytics with spatial cropping, enhanced uniformerv2 and modified efficient additive attention.arXiv e-prints, pages arXiv–2404, 2024

    Damith Chamalke Senadeera, Xiaoyun Yang, Dimitrios Kol- lias, and Gregory Slabaugh. Cue-net: Violence detection video analytics with spatial cropping, enhanced uniformerv2 and modified efficient additive attention.arXiv e-prints, pages arXiv–2404, 2024. 6

    2024

  2. [11]

    Rwf-2000: an open large scale video database for violence detection

    Ming Cheng, Kunjing Cai, and Ming Li. Rwf-2000: an open large scale video database for violence detection. In 2020 25th International Conference on Pattern Recognition (ICPR), pages 4183–4190. IEEE, 2021. 1, 2, 3

    2000

  3. [12]

    Covid- 19 computer-aided diagnosis through ai-assisted ct imaging analysis: Deploying a medical ai system.arXiv preprint arXiv:2403.06242, 2024

    Demetris Gerogiannis, Anastasios Arsenos, Dimitrios Kol- lias, Dimitris Nikitopoulos, and Stefanos Kollias. Covid- 19 computer-aided diagnosis through ai-assisted ct imaging analysis: Deploying a medical ai system.arXiv preprint arXiv:2403.06242, 2024. 4 7

    2024 arXiv

  4. [13]

    Violent flows: Real-time detection of violent crowd behavior

    Tal Hassner, Yossi Itcher, and Orit Kliper-Gross. Violent flows: Real-time detection of violent crowd behavior. In 2012 IEEE Computer Society Conference on Computer Vi- sion and Pattern Recognition Workshops, pages 1–6, 2012. 2

    2012

  5. [14]

    Rethinking affect analysis: A protocol for ensuring fairness and consis- tency.arXiv preprint arXiv:2408.02164, 2024

    Guanyu Hu, Dimitrios Kollias, Eleni Papadopoulou, Paraskevi Tzouveli, Jie Wei, and Xinyu Yang. Rethinking affect analysis: A protocol for ensuring fairness and consis- tency.arXiv preprint arXiv:2408.02164, 2024. 4

    2024 arXiv

  6. [15]

    Bridging the gap: Protocol towards fair and consistent affect analysis

    Guanyu Hu, Eleni Papadopoulou, Dimitrios Kollias, Paraskevi Tzouveli, Jie Wei, and Xinyu Yang. Bridging the gap: Protocol towards fair and consistent affect analysis. arXiv preprint arXiv:2405.06841, 2024

    2024 arXiv

  7. [16]

    Ensuring uav safety: A vision-only and real-time framework for collision avoidance through object detection, tracking, and distance estimation.arXiv preprint arXiv:2405.06749, 2024

    Vasileios Karampinis, Anastasios Arsenos, Orfeas Filip- popoulos, Evangelos Petrongonas, Christos Skliros, Dim- itrios Kollias, Stefanos Kollias, and Athanasios V oulodimos. Ensuring uav safety: A vision-only and real-time framework for collision avoidance through object detec...

    2024 arXiv

  8. [17]

    Abaw: Valence-arousal estimation, ex- pression recognition, action unit detection & multi-task learning challenges.arXiv preprint arXiv:2202.10659, 2022

    Dimitrios Kollias. Abaw: Valence-arousal estimation, ex- pression recognition, action unit detection & multi-task learning challenges.arXiv preprint arXiv:2202.10659, 2022

    2022 arXiv

  9. [18]

    Multi-label compound expression recog- nition: C-expr database & network

    Dimitrios Kollias. Multi-label compound expression recog- nition: C-expr database & network. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5589–5598, 2023. 4

    2023

  10. [19]

    Aff-wild2: Ex- tending the aff-wild database for affect recognition.arXiv preprint arXiv:1811.07770, 2018

    Dimitrios Kollias and Stefanos Zafeiriou. Aff-wild2: Ex- tending the aff-wild database for affect recognition.arXiv preprint arXiv:1811.07770, 2018

    2018 arXiv

  11. [20]

    A multi-task learn- ing & generation framework: Valence-arousal, action units & primary expressions.arXiv preprint arXiv:1811.07771, 2018

    Dimitrios Kollias and Stefanos Zafeiriou. A multi-task learn- ing & generation framework: Valence-arousal, action units & primary expressions.arXiv preprint arXiv:1811.07771, 2018

    2018 arXiv

  12. [21]

    A multi- component cnn-rnn approach for dimensional emotion recognition in-the-wild.arXiv preprint arXiv:1805.01452, 2018

    Dimitrios Kollias and Stefanos Zafeiriou. A multi- component cnn-rnn approach for dimensional emotion recognition in-the-wild.arXiv preprint arXiv:1805.01452, 2018

    2018 arXiv

  13. [22]

    Training deep neu- ral networks with different datasets in-the-wild: The emotion recognition paradigm

    Dimitrios Kollias and Stefanos Zafeiriou. Training deep neu- ral networks with different datasets in-the-wild: The emotion recognition paradigm. In2018 International Joint Confer- ence on Neural Networks (IJCNN), pages 1–8. IEEE, 2018

    2018

  14. [23]

    Expression, affect, action unit recognition: Aff-wild2, multi-task learning and arcface.arXiv preprint arXiv:1910.04855, 2019

    Dimitrios Kollias and Stefanos Zafeiriou. Expression, affect, action unit recognition: Aff-wild2, multi-task learning and arcface.arXiv preprint arXiv:1910.04855, 2019

    1910 arXiv

  15. [24]

    Va-stargan: Con- tinuous affect generation

    Dimitrios Kollias and Stefanos Zafeiriou. Va-stargan: Con- tinuous affect generation. InInternational Conference on Advanced Concepts for Intelligent Vision Systems, pages 227–238. Springer, 2020

    2020

  16. [25]

    Affect analysis in-the-wild: Valence-arousal, expressions, action units and a unified framework.arXiv preprint arXiv:2103.15792, 2021

    Dimitrios Kollias and Stefanos Zafeiriou. Affect analysis in-the-wild: Valence-arousal, expressions, action units and a unified framework.arXiv preprint arXiv:2103.15792, 2021

    2021 arXiv

  17. [26]

    Analysing affec- tive behavior in the second abaw2 competition

    Dimitrios Kollias and Stefanos Zafeiriou. Analysing affec- tive behavior in the second abaw2 competition. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 3652–3660, 2021

    2021

  18. [27]

    Exploiting multi-cnn features in cnn-rnn based dimensional emotion recognition on the omg in-the-wild dataset.IEEE Transac- tions on Affective Computing, 2020

    Dimitrios Kollias and Stefanos P Zafeiriou. Exploiting multi-cnn features in cnn-rnn based dimensional emotion recognition on the omg in-the-wild dataset.IEEE Transac- tions on Affective Computing, 2020

    2020

  19. [28]

    Analysing affective behavior in the first abaw 2020 competition

    D Kollias, A Schulc, E Hajiyev, and S Zafeiriou. Analysing affective behavior in the first abaw 2020 competition. In 2020 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020)(FG), pages 794– 800,

    2020

  20. [29]

    Advancements in affective and behavior analysis: The 8th abaw workshop and competition

    Dimitrios Kollias, Panagiotis Tzirakis, Alan Cowen, Irene Kotsia, UK Cogitat, Eric Granger, Marco Pedersoli, Simon Bacon, Alice Baird, Chunchang Shao, et al. Advancements in affective and behavior analysis: The 8th abaw workshop and competition

  21. [30]

    Inter- weaving deep learning and semantic techniques for emotion analysis in human-machine interaction

    Dimitris Kollias, George Marandianos, Amaryllis Raouzaiou, and Andreas-Georgios Stafylopatis. Inter- weaving deep learning and semantic techniques for emotion analysis in human-machine interaction. In2015 10th International Workshop on Semantic and Social Media Adaptation and ...

    2015

  22. [31]

    On line emotion detection using retrainable deep neural networks

    Dimitrios Kollias, Athanasios Tagaris, and Andreas Stafy- lopatis. On line emotion detection using retrainable deep neural networks. InComputational Intelligence (SSCI), 2016 IEEE Symposium Series on, pages 1–8. IEEE, 2016

    2016

  23. [32]

    Recognition of affect in the wild using deep neural networks

    Dimitrios Kollias, Mihalis A Nicolaou, Irene Kotsia, Guoy- ing Zhao, and Stefanos Zafeiriou. Recognition of affect in the wild using deep neural networks. InComputer Vision and Pattern Recognition Workshops (CVPRW), 2017 IEEE Conference on, pages 1972–1979. IEEE, 2017

    2017

  24. [33]

    Adap- tation and contextualization of deep neural network models

    Dimitrios Kollias, Miao Yu, Athanasios Tagaris, Georgios Leontidis, Andreas Stafylopatis, and Stefanos Kollias. Adap- tation and contextualization of deep neural network models. InComputational Intelligence (SSCI), 2017 IEEE Sympo- sium Series on, pages 1–8. IEEE, 2017

    2017

  25. [34]

    Photorealistic facial synthesis in the dimensional affect space

    Dimitrios Kollias, Shiyang Cheng, Maja Pantic, and Stefanos Zafeiriou. Photorealistic facial synthesis in the dimensional affect space. InProceedings of the European Conference on Computer Vision (ECCV) Workshops, pages 0–0, 2018

    2018

  26. [35]

    Deep neural archi- tectures for prediction in healthcare.Complex & Intelligent Systems, 4(2):119–131, 2018

    Dimitrios Kollias, Athanasios Tagaris, Andreas Stafylopatis, Stefanos Kollias, and Georgios Tagaris. Deep neural archi- tectures for prediction in healthcare.Complex & Intelligent Systems, 4(2):119–131, 2018

    2018

  27. [36]

    Face behavior a la carte: Expressions, af- fect and action units in a single network.arXiv preprint arXiv:1910.11111, 2019

    Dimitrios Kollias, Viktoriia Sharmanska, and Stefanos Zafeiriou. Face behavior a la carte: Expressions, af- fect and action units in a single network.arXiv preprint arXiv:1910.11111, 2019. 4

    1910 arXiv

  28. [37]

    Deep affect prediction in-the-wild: Aff-wild database and challenge, deep architec- tures, and beyond.International Journal of Computer Vision, 127(6):907–929, 2019

    Dimitrios Kollias, Panagiotis Tzirakis, Mihalis A Nicolaou, Athanasios Papaioannou, Guoying Zhao, Bj ¨orn Schuller, Irene Kotsia, and Stefanos Zafeiriou. Deep affect prediction in-the-wild: Aff-wild database and challenge, deep architec- tures, and beyond.International Journal...

    2019

  29. [38]

    Deep transparent prediction through latent represen- tation analysis.arXiv preprint arXiv:2009.07044, 2020

    Dimitrios Kollias, N Bouas, Y Vlaxos, V Brillakis, M Se- feris, Ilianna Kollia, Levon Sukissian, James Wingate, and S Kollias. Deep transparent prediction through latent represen- tation analysis.arXiv preprint arXiv:2009.07044, 2020. 8

    2009 arXiv

  30. [39]

    Deep neural network augmentation: Generating faces for affect analysis.Interna- tional Journal of Computer Vision, pages 1–30, 2020

    Dimitrios Kollias, Shiyang Cheng, Evangelos Ververas, Irene Kotsia, and Stefanos Zafeiriou. Deep neural network augmentation: Generating faces for affect analysis.Interna- tional Journal of Computer Vision, pages 1–30, 2020

    2020

  31. [40]

    Transpar- ent adaptation in deep medical image diagnosis

    Dimitris Kollias, Y Vlaxos, M Seferis, Ilianna Kollia, Levon Sukissian, James Wingate, and Stefanos D Kollias. Transpar- ent adaptation in deep medical image diagnosis. InTAILOR, pages 251–267, 2020

    2020

  32. [41]

    Distribution matching for heterogeneous multi- task learning: a large-scale face study.arXiv preprint arXiv:2105.03790, 2021

    Dimitrios Kollias, Viktoriia Sharmanska, and Stefanos Zafeiriou. Distribution matching for heterogeneous multi- task learning: a large-scale face study.arXiv preprint arXiv:2105.03790, 2021. 4

    2021 arXiv

  33. [42]

    Ai-enabled analysis of 3-d ct scans for diagnosis of covid-19 & its severity

    Dimitrios Kollias, Anastasios Arsenos, and Stefanos Kollias. Ai-enabled analysis of 3-d ct scans for diagnosis of covid-19 & its severity. In2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICAS- SPW), pages 1–5. IEEE, 2023

    2023

  34. [43]

    A deep neural architecture for harmonizing 3-d input data analysis and decision making in medical imaging.Neuro- computing, 542:126244, 2023

    Dimitrios Kollias, Anastasios Arsenos, and Stefanos Kollias. A deep neural architecture for harmonizing 3-d input data analysis and decision making in medical imaging.Neuro- computing, 542:126244, 2023

    2023

  35. [44]

    Ai-mia: Covid-19 detection and severity analysis through medical imaging

    Dimitrios Kollias, Anastasios Arsenos, and Stefanos Kollias. Ai-mia: Covid-19 detection and severity analysis through medical imaging. InComputer Vision–ECCV 2022 Work- shops: Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part VII, pages 677–690. Springer, 2023

    2022

  36. [46]

    Mma-mrnnet: Harnessing multi- ple models of affect and dynamic masked rnn for pre- cise facial expression intensity estimation.arXiv preprint arXiv:2303.00180, 2023

    Dimitrios Kollias, Andreas Psaroudakis, Anastasios Ar- senos, Paraskevi Theofilou, Chunchang Shao, Guanyu Hu, and Ioannis Patras. Mma-mrnnet: Harnessing multi- ple models of affect and dynamic masked rnn for pre- cise facial expression intensity estimation.arXiv preprint arXiv...

    2023 arXiv

  37. [47]

    Abaw: Valence-arousal esti- mation, expression recognition, action unit detection & emo- tional reaction intensity estimation challenges

    Dimitrios Kollias, Panagiotis Tzirakis, Alice Baird, Alan Cowen, and Stefanos Zafeiriou. Abaw: Valence-arousal esti- mation, expression recognition, action unit detection & emo- tional reaction intensity estimation challenges. InProceed- ings of the IEEE/CVF Conference on Comp...

    2023

  38. [48]

    Btdnet: A multi-modal approach for brain tumor radiogenomic classification.Applied Sciences, 13(21):11984, 2023

    Dimitrios Kollias, Karanjot Vendal, Priyankaben Gadhavi, and Solomon Russom. Btdnet: A multi-modal approach for brain tumor radiogenomic classification.Applied Sciences, 13(21):11984, 2023

    2023

  39. [49]

    Domain adaptation, explainability & fairness in ai for medi- cal image analysis: Diagnosis of covid-19 based on 3-d chest ct-scans.arXiv preprint arXiv:2403.02192, 2024

    Dimitrios Kollias, Anastasios Arsenos, and Stefanos Kollias. Domain adaptation, explainability & fairness in ai for medi- cal image analysis: Diagnosis of covid-19 based on 3-d chest ct-scans.arXiv preprint arXiv:2403.02192, 2024

    2024 arXiv

  40. [50]

    Sam2clip2sam: Vision language model for segmentation of 3d ct scans for covid-19 detection.arXiv preprint arXiv:2407.15728, 2024

    Dimitrios Kollias, Anastasios Arsenos, James Wingate, and Stefanos Kollias. Sam2clip2sam: Vision language model for segmentation of 3d ct scans for covid-19 detection.arXiv preprint arXiv:2407.15728, 2024

    2024 arXiv

  41. [51]

    Behaviour4all: in-the-wild facial behaviour analysis toolkit.arXiv preprint arXiv:2409.17717, 2024

    Dimitrios Kollias, Chunchang Shao, Odysseus Kaloidas, and Ioannis Patras. Behaviour4all: in-the-wild facial behaviour analysis toolkit.arXiv preprint arXiv:2409.17717, 2024

    2024 arXiv

  42. [52]

    Distribution matching for multi-task learning of classification tasks: a large-scale study on faces & beyond

    Dimitrios Kollias, Viktoriia Sharmanska, and Stefanos Zafeiriou. Distribution matching for multi-task learning of classification tasks: a large-scale study on faces & beyond. arXiv preprint arXiv:2401.01219, 2024. 4

    2024 arXiv

  43. [53]

    The 6th affective behavior analysis in-the-wild (abaw) competition

    Dimitrios Kollias, Panagiotis Tzirakis, Alan Cowen, Ste- fanos Zafeiriou, Chunchang Shao, and Guanyu Hu. The 6th affective behavior analysis in-the-wild (abaw) competition. arXiv preprint arXiv:2402.19344, 2024

    2024 arXiv

  44. [54]

    7th abaw competition: Multi-task learning and compound expression recognition.arXiv preprint arXiv:2407.03835, 2024

    Dimitrios Kollias, Stefanos Zafeiriou, Irene Kotsia, Abhinav Dhall, Shreya Ghosh, Chunchang Shao, and Guanyu Hu. 7th abaw competition: Multi-task learning and compound expression recognition.arXiv preprint arXiv:2407.03835, 2024

    2024 arXiv

  45. [55]

    Cowen, Ste- fanos Zafeiriou, Irene Kotsia, Eric Granger, Marco Peder- soli, Simon L

    Dimitrios Kollias, Panagiotis Tzirakis, Alan S. Cowen, Ste- fanos Zafeiriou, Irene Kotsia, Eric Granger, Marco Peder- soli, Simon L. Bacon, Alice Baird, Chris Gagne, Chun- chang Shao, Guanyu Hu, Soufiane Belharbi, and Muham- mad Haseeb Aslam. Advancements in Affective and Beha...

    2025

  46. [56]

    Can machine learning assist in diagnosis of primary immune thrombocytopenia? a feasibility study.arXiv preprint arXiv:2405.20562, 2024

    Haroon Miah, Dimitrios Kollias, Giacinto Luca Pedone, Drew Provan, and Frederick Chen. Can machine learning assist in diagnosis of primary immune thrombocytopenia? a feasibility study.arXiv preprint arXiv:2405.20562, 2024. 4

    2024 arXiv

  47. [57]

    A multi-stream cnn for deep violence detection in video sequences using handcrafted features.The Visual Computer, 38(6):2057–2072, 2022

    Seyed Mehdi Mohtavipour, Mahmoud Saeidi, and Abouzar Arabsorkhi. A multi-stream cnn for deep violence detection in video sequences using handcrafted features.The Visual Computer, 38(6):2057–2072, 2022. 3

    2022

  48. [58]

    Movies fight detection dataset

    Enrique Bermejo Nievas, Oscar Deniz Suarez, Gloria Bueno Garcia, and Rahul Sukthankar. Movies fight detection dataset. InComputer Analysis of Images and Patterns, pages 332–339. Springer, 2011. 2, 3

    2011

  49. [59]

    Mixaugment & mixup: Augmentation methods for facial expression recog- nition

    Andreas Psaroudakis and Dimitrios Kollias. Mixaugment & mixup: Augmentation methods for facial expression recog- nition. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 2367–2375,

  50. [60]

    Medical image segmentation: A review of modern architec- tures

    Natalia Salpea, Paraskevi Tzouveli, and Dimitrios Kollias. Medical image segmentation: A review of modern architec- tures. InEuropean Conference on Computer Vision, pages 691–708. Springer, 2022. 4

    2022

  51. [61]

    Cue-net: violence detection video analytics with spatial cropping enhanced uniformerv2 and modified efficient additive attention

    Damith Chamalke Senadeera, Xiaoyun Yang, Dimitrios Kol- lias, and Gregory Slabaugh. Cue-net: violence detection video analytics with spatial cropping enhanced uniformerv2 and modified efficient additive attention. InProceedings of the IEEE/CVF Conference on Computer Vision and...

    2024

  52. [62]

    Violence recog- nition from videos using deep learning techniques

    Mohamed Mostafa Soliman, Mohamed Hussein Kamal, Mina Abd El-Massih Nashed, Youssef Mohamed Mostafa, Bassel Safwat Chawky, and Dina Khattab. Violence recog- nition from videos using deep learning techniques. In2019 Ninth International Conference on Intelligent Computing and Inf...

    2019

  53. [63]

    Violence recog- nition from videos using deep learning techniques

    Mohamed Mostafa Soliman, Mohamed Hussein Kamal, Mina Abd El-Massih Nashed, Youssef Mohamed Mostafa, Bassel Safwat Chawky, and Dina Khattab. Violence recog- nition from videos using deep learning techniques. In 9 2019 Ninth International Conference on Intelligent Comput- ing an...

    2019

  54. [64]

    Real-world anomaly detection in surveillance videos

    Waqas Sultani, Chen Chen, and Mubarak Shah. Real-world anomaly detection in surveillance videos. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6479–6488, 2018. 2

    2018

  55. [65]

    Assessment of parkinson’s disease based on deep neural networks

    Athanasios Tagaris, Dimitrios Kollias, and Andreas Stafy- lopatis. Assessment of parkinson’s disease based on deep neural networks. InInternational Conference on Engi- neering Applications of Neural Networks, pages 391–403. Springer, 2017. 4

    2017

  56. [66]

    Athanasios Tagaris, Dimitrios Kollias, Andreas Stafylopatis, Georgios Tagaris, and Stefanos Kollias. Machine learning for neurodegenerative disorder diagnosis—survey of prac- tices and launch of benchmark dataset.International Journal on Artificial Intelligence Tools, 27(03):1...

    2018

  57. [67]

    Not only look, but also listen: Learning multimodal violence detection under weak supervision

    Peng Wu, Jing Liu, Yujia Shi, Yujia Sun, Fangtao Shao, Zhaoyang Wu, and Zhiwei Yang. Not only look, but also listen: Learning multimodal violence detection under weak supervision. InComputer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Pro- ce...

    2020

  58. [68]

    Video violence rating: A large-scale public database and a multimodal rating model.IEEE Transactions on Multimedia, 26:8557–8568,

    Tao Xiang, Hongyan Pan, and Zhixiong Nan. Video violence rating: A large-scale public database and a multimodal rating model.IEEE Transactions on Multimedia, 26:8557–8568,

  59. [69]

    Berg, and Dimitris Samaras

    Kiwon Yun, Jean Honorio, Debaleena Chattopadhyay, Tamara L. Berg, and Dimitris Samaras. Two-person interac- tion detection using body-pose features and multiple instance learning. In2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, pag...

    2012

  60. [70]

    Aff-wild: Valence and arousal ‘in-the-wild’challenge

    Stefanos Zafeiriou, Dimitrios Kollias, Mihalis A Nicolaou, Athanasios Papaioannou, Guoying Zhao, and Irene Kot- sia. Aff-wild: Valence and arousal ‘in-the-wild’challenge. InComputer Vision and Pattern Recognition Workshops (CVPRW), 2017 IEEE Conference on, pages 1980–1987. IEE...

    2017

Pith tools

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