Pith. sign in

REVIEW 3 major objections 4 minor 43 references

IDraw: Artist Verification from Digital Drawing Images

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read IDraw verifies digital-drawing authorship from images alone, cutting verification error by up to 40%.

desk verdict A real dataset and a sensible privileged-supervision pipeline with consistent gains, but the artist-device confound and loose statistics need work before I'd trust the behavioral interpretation. read the letter →

arxiv 2608.01737 v2 pith:GE4CE5E7 submitted 2026-08-03 cs.CV

Nayoung Kim , Nan Jiang , Bangjie Sun , Jaewon Shin , Sojeong Kim , Jun Han This is my paper · ORCID
classification cs.CV
keywords drawingauthorshipverificationartisttablet-pensensorsignalsbehavior-guidedlearningcontentsuppressionmultimodaldatasetdigitaldrawingsusingprivilegedinformation
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

This paper claims that artist-specific drawing behavior—pen pressure, stroke timing, editing habits—leaves recoverable traces in a finished digital drawing, and that a verification system can be trained to read those traces even when the claimed artist's sensor data is never available. The proposed framework, IDraw, trains an image encoder on drawings paired with 14 tablet-pen sensor signals collected from separate artists, teaching the encoder to make embedding differences mirror behavior differences. At verification time only the disputed drawing and trusted reference drawings are used, with an added step that subtracts an estimated object mean to suppress what the drawing depicts. On a newly collected dataset of 1,110 drawings by 37 artists, each drawing all 30 objects, IDraw consistently outperforms standard image-based verification across nine encoder backbones and all six reference-set sizes, reducing verification error by up to 40 percent.

What carries the argument

The machinery has two load-bearing parts. First, behavior-guided supervision (M3): a linear head $A$ maps the difference $\Delta \hat{z}$ between two normalized image embeddings to the difference $\Delta \tilde{b}$ between their standardized behavior feature vectors, trained by the mean-squared-error loss $L_{\text{beh}} = \frac{1}{K}\|A\Delta\hat{z} - \Delta\tilde{b}\|_2^2$; this is what teaches the image encoder to read behavior from pixels. Second, content suppression (M5): for each object $o$, the mean embedding $\mu_o$ over training artists is subtracted from every drawing's standardized embedding before $\ell^2$ normalization, removing the component shared by same-object drawings. A triplet loss (M4) with batch-hard mining pulls same-artist different-object drawings together and pushes same-object different-artist drawings apart. The verification score is the cosine distance between the content-suppressed query representation and the $\ell^2$-normalized average of content-suppressed reference representations.

What would settle it

A cross-device experiment: recruit the same artists to draw the same objects on two different tablet models, train IDraw on artists using device A, and verify on artists using device B; if AUC falls to near baseline while same-device verification stays high, device-specific pen artifacts rather than artist behavior are carrying the signal. A quicker check is whether the 50 behavior features for one artist change more across devices than across artists.

Watch

Extended reading notes

Core claim

IDraw's central claim is that a completed digital drawing retains enough of the artist's production behavior—pressure, stroke timing, revision patterns—that a model trained with tablet-pen sensor data from other artists can verify authorship from images alone. The key design choice is to supervise the difference between normalized image embeddings of two same-object drawings by different artists with the difference between their standardized 50-dimensional behavior feature vectors, so the embedding geometry is pushed to track behavior rather than object appearance. At verification, the method subtracts an estimated object mean from each embedding and averages the reference drawings to form an artist representation; the cosine distance to the query drawing is the verification score. In the authors' evaluation on 37 artists and 30 objects, this recipe outperforms the cross-entropy baseline on all nine encoders and all six reference-set sizes, with relative equal-error-rate reductions up to 40 percent, and it closes 58.1 percent of the gap between baseline and the sensor-signal upper bound on one representative encoder.

Load-bearing premise

The load-bearing premise is that pen-pressure, timing, and tilt signals collected on each artist's own tablet and digital pen are comparable across devices and reflect drawing behavior, not device-specific artifacts; if that fails, the sensor supervision teaches device fingerprints and the measured gains would not mean what the paper claims.

Editorial extensions

If this is right

  • Reference drawings need not depict the same object as the disputed drawing, so verification can work across an artist's varied output.
  • No sensor data is needed from the claimed artist at verification time; the sensor guidance is baked into the image encoder during training, so the method applies wherever finished images are shared.
  • The gains appear across nine image encoders spanning different pretraining paradigms, suggesting the behavior signal is broadly available rather than tied to one architecture.
  • The fully crossed dataset (37 artists, 30 objects, 14 sensor signals) enables controlled study of artist versus object variation that existing drawing datasets cannot support.

Reading between the lines

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

  • A natural extension the paper leaves implicit: resizing, cropping, or JPEG compression may destroy some behavior traces, so testing IDraw on redistributed images would reveal whether the gains survive real online sharing.
  • Because the object mean is estimated from only 23 training artists, the method could be made more robust by estimating content statistics from a larger artist pool or from unlabeled same-object drawings available at verification time.
  • If generative models imitate visual style but not pen-pressure and timing statistics, behavior-guided verification could serve as a complementary defense against style imitation; the paper names this as an open question rather than a result.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes IDraw, a framework for verifying the authorship of digital drawings from completed images only. During training, IDraw uses tablet-pen sensor signals recorded from separate training artists as privileged supervision: an image encoder is trained with a pairwise behavior loss (Eq. 2) that regresses differences in standardized sensor-derived behavior features from differences in normalized image embeddings, and a triplet loss (Eq. 3) that pulls same-artist different-object drawings together and pushes same-object different-artist drawings apart. At verification time, the method subtracts per-object mean embeddings estimated from training artists (Eq. 5), aggregates reference drawings, and uses cosine distance to the query. The authors collect a fully crossed dataset of 1,110 drawings from 37 artists drawing 30 objects, each paired with 14 types of tablet-pen sensor signals. They evaluate on unseen artists across nine image encoders and six reference-set sizes, reporting consistent AUC/EER gains over a cross-entropy baseline, including a reduction in verification error of up to 40% and an average AUC gain of 0.138 at N=1. Ablations show that behavior-guided supervision, triplet loss, and content suppression each contribute to the gains.

Significance. If the results hold, the paper makes a useful contribution: it demonstrates a learning-using-privileged-information approach to offline drawing verification and introduces a dataset that is, to my knowledge, the first fully crossed artist-object drawing dataset with tablet-pen sensor signals. The experimental sweep over nine encoders and multiple reference-set sizes is unusually thorough, and the module-level ablations (Tables 3 and 4) provide a clear picture of where the gains come from. The main contributions are potentially valuable for practical authorship disputes where only completed images are available. However, the central behavioral interpretation rests on the assumption that the sensor-signal supervision is indicative of artist-specific drawing behavior rather than device-specific artifacts; the current manuscript does not provide the control experiments needed to establish this. The statistical significance claims are also overstated because the 15 splits are not independent. The work is significant but needs substantial strengthening before its core claim can be accepted.

major comments (3)
  1. [Data Collection; Eq. (2)] The sensor signals are recorded from each participant's own iPad and Apple Pencil, so device identity is completely confounded with artist identity in the training data. The 50 selected behavior features include pressure, timing, and stroke geometry, all of which are plausibly affected by device model, Pencil generation, pressure-curve calibration, sampling rate, and rendering pipeline. Because the completed drawing images are rendered and captured on the same device, the image encoder can learn visual fingerprints of the device (e.g., resolution, line rasterization, color profile) that correlate with the sensor features. The M3 regression in Eq. (2) can then be satisfied by encoding device-specific rendering differences rather than motor behavior. The paper reports no device-type metadata, no per-device analysis, and no cross-device evaluation. This is load-bearing because the paper's central claim is that IDraw infers behavior-related artist cues and will generalize to previously unseen artists; if the gains instead reflect device fingerprints, the behavioral interpretation and the generalization claim collapse. A minimum fix would be a leave-one-device-out or cross-device experiment, or at least a demonstration that the gains persist when device information is controlled.
  2. [Evaluation Protocol; Overall Results] The significance testing is not valid as reported. The 15 artist-disjoint splits are not independent: each split selects 23 fit, 6 validation, and 8 verification artists from the same fixed pool of 37 artists, so the same artists and drawings appear across splits. Paired t-tests across these 15 splits therefore treat 15 dependent measurements as independent, and the reported p-values below 10^-6 (for all 54 combinations) overstate the statistical evidence. In addition, all experiments use a single training seed. The central claim that IDraw consistently outperforms the baseline would be much better supported by reporting variance across multiple seeds or by using an artist-level bootstrap that respects the resampling structure. This is a load-bearing point because the paper's 'consistent' improvement is asserted largely through these significance claims.
  3. [Content Suppression, Eq. (5)] The object mean mu_o in Eq. (5) is estimated from the 23 training artists and is then subtracted from embeddings of unseen verification artists. This presumes that the per-object mean of the training artists is a good estimate of the artist-independent content component for the verification artists. If the training artists have systematic style correlations, or if object identity and artist identity are not fully separable, subtracting this mean may either fail to remove content or may remove artist-specific identity cues. The paper reports only a single instantiation of the dataset split, so there is no evidence about the stability of mu_o across different training-artist subsets. A simple experiment would estimate mu_o from random halves of the training artists and compare verification performance; the current ablation (Table 4, 'Content Suppress.') is not sufficient to establish the robustness of this central mechanism.
minor comments (4)
  1. [Table 2] The table reports five of the nine encoders, yet the text says that all nine encoders improve at every N and cites 'supplementary material' for full results. The main text should either include all nine encoders or clearly state that the five shown are representative and provide a complete table in an appendix that is accessible to the reader.
  2. [Table 1] The 'Fully crossed' column header is ambiguous: the check mark in the IDraw row should be defined explicitly in the caption as 'every artist draws every object' to avoid confusion with the 'crossed' design terminology used in statistics.
  3. [Figure 4(a)] The improvement in behavior-feature prediction correlations is small (mean gain +0.016 over 50 features) and the figure shows overlapping error bars. The text claims 'consistent' improvement, but it would be helpful to state the effect size in standard deviations and to show per-feature confidence intervals for the key families (timing, pressure, revision).
  4. [Dataset Availability] The dataset and code are not released. Given that the paper introduces a new dataset and a new training procedure, releasing these artifacts would be important for reproducibility and for verifying that the device-confounding concern is not material. At minimum, the authors should state a release plan or provide a detailed description of the sensor-signal preprocessing and feature definitions in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the behavioral supervision target and content-suppression means are estimated from training artists only, and all headline results are evaluated on held-out verification artists.

full rationale

The paper's central derivation is self-contained and does not reduce to its inputs. The behavior loss in Eq. (2) regresses differences in standardized sensor-derived behavior features (Delta b) from differences in normalized image embeddings (Delta z), but Delta b is built from tablet-pen sensor signals recorded during the training artists' drawing sessions (M2, Data Collection), not from the verification artists' identities or from the evaluation labels. Verification is performed on 8 artists per split who are disjoint from the 23 fit and 6 validation artists, and every headline number in Table 2 is computed on those held-out verification artists; no verification query or reference drawing contributes to training the encoder or to Eq. (5). The object means in Eq. (5) are averaged over Atrain only, so content suppression uses no target-artist information. The only tuning choices, the behavior-loss weight lambda and training epoch, are selected on validation artists and are standard model selection rather than a fitted prediction. The self-citations to CAMPrints (Sun, Chan, and Han 2025), UniKey (Tan, Chan, and Han 2025), and Han et al. (2018) appear only as related-work examples of cross-modal sensing and privileged information; none is invoked as a uniqueness theorem or as a load-bearing premise that forces the IDraw design. The weakest assumptions, such as cross-device comparability of the collected sensor signals, are empirical confound concerns and not circularity: they question whether the sensor signal encodes behavior or device artifacts, but they do not make any claimed prediction equivalent to its training input by construction.

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

The central model uses one tuned scalar, a fixed triplet margin, and a hand-constructed 50-feature behavior vector. No new physical or conceptual entities are postulated. The object mean is a data statistic, not a free parameter, but the assumption that it is artist-independent is listed as an axiom.

free parameters (4)
  • behavior-loss weight lambda = per-encoder value selected on validation AUC, not reported
    Balances L_beh and L_tri in Eq. (4). Selection on validation artists means the reported gains partly reflect validation tuning.
  • triplet margin m = 0.2
    Fixed cosine margin in Eq. (3); not swept, so it is a minor free choice.
  • embedding dimension = 256
    Output size of projection head in Eq. (1); architectural choice affecting capacity.
  • behavior feature set size K = 50
    Reduced from 109 candidates by removing image-computable features, near-duplicates with Spearman >= 0.95, and nine tilt features. These selection rules are hand-chosen and determine what M3 can supervise.
assumptions (4)
  • domain assumption The 50 handcrafted sensor features are artist-discriminative and summarize drawing behavior well enough to guide image learning.
    M2 discards all raw signals into K=50 statistics; if the statistics omit the identity signal, M3 cannot recover it.
  • domain assumption Same-object drawings by different artists share an artist-independent content component that a training-artist mean can estimate and subtract.
    Eq. (5) subtracts the object mean from unseen verification artists; if content varies by artist, this step could remove identity information.
  • domain assumption Verification artists are drawn from the same population as training artists for both behavior and object content.
    The 15 splits assume transfer from 23 fit artists to 8 verification artists under stationary conditions.
  • domain assumption Personal iPads and Apple Pencils produce sensor signals that are comparable across devices and reflect motor behavior, not device calibration.
    Data Collection uses each participant's own device; no device normalization or control experiment is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IDraw: Artist Verification from Digital Drawing Images." pith.science (2026). https://pith.science/paper/GE4CE5E7

@misc{pith2026260801737,
  author       = {Pith},
  title        = {Pith review of: IDraw: Artist Verification from Digital Drawing Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GE4CE5E7}},
  note         = {Machine review of arXiv:2608.01737}
}
read the original abstract

As digital drawings are increasingly shared online, reliable authorship verification has become important for protecting artists and resolving disputes. Yet when authorship is questioned, verification may have to rely only on the disputed drawing and reference drawings known to be created by the claimed artist. This setting is challenging for two reasons. First, artist-specific drawing behavior, such as pen pressure and movement speed, is informative but is not available from a completed drawing. Second, similarities in the depicted object or scene can obscure similarities arising from the artist. We propose IDraw, a framework that learns from drawings paired with tablet-pen sensor signals collected from separate training artists. This allows IDraw to infer drawing behavior from completed images during a later authorship dispute, without requiring sensor data from the artist being verified. IDraw also reduces the influence of drawing content by identifying information shared by drawings of the same object across different artists and suppressing it before comparing drawings. To support this approach, we construct the first multimodal dataset for digital drawing authorship verification, containing 1,110 drawings from 37 artists and 14 types of tablet-pen sensor signals. Evaluated on previously unseen artists across nine image-encoder backbones, IDraw consistently outperforms standard image-based verification and reduces verification error by up to 40%. These results demonstrate that inferring drawing behavior from completed images and suppressing drawing content improve digital drawing authorship verification.

Figures

Figures reproduced from arXiv: 2608.01737 by the authors.

Figure 1
Figure 1. Overview of IDraw. (a) IDraw verifies whether a disputed drawing is created by the claimed artist using the artist’s reference drawings. (b) Training phase: Modules M1 through M4 learn an image embedding guided by tablet-pen sensor signals and organized by artist identity. Verification phase: Modules M5 through M7 use only completed drawings to suppress remaining drawing content, aggregate the reference drawings, an… view at source ↗
Figure 2
Figure 2. Behavior-guided supervision (M3). (a) A pair of same-object drawings by different artists has a small im￾age embedding difference. (b) The same drawing pair has a large difference in drawing behavior. (c) M3 makes the im￾age embedding difference (∆zˆ) reflect the drawing behavior difference (∆b˜). image embedding. M2 converts its paired tablet-pen sensor signals into a fixed-size drawing behavior feature vector. M3 … view at source ↗
Figure 3
Figure 3. Summary of the IDraw dataset. Each cell of the 37×30 matrix is a participant’s drawing of the visual prompt, paired with the 14 tablet-pen sensor signals recorded during its creation. Data Collection. Participants use Apple Pencils on their own Apple iPads with an open-source tool (Ding 2016). Each participant draws the same 30 objects using visual prompts, while the tool recorded 14 tablet-pen sensor signals, inclu… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: (a) compares behavior feature prediction for full IDraw and IDraw without behavior-guided supervision (M3) on DINOv2-B. We group the 50 behavior features into five families. Without behavior-guided supervision, prediction correlations range from approximately r = 0.18 …

Review history (3 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 40 canonical work pages

  1. [1]

    StegaStamp: Invisible Hyperlinks in Physical Photographs

    Tancik, Matthew and Mildenhall, Ben and Ng, Ren. StegaStamp: Invisible Hyperlinks in Physical Photographs. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    work page

  2. [2]

    Invisible Image Watermarks Are Provably Removable Using Generative AI

    Zhao, Xuandong and Zhang, Kexun and Su, Zihao and Vasan, Saastha and Grishchenko, Ilya and Kruegel, Christopher and Vigna, Giovanni and Wang, Yu-Xiang and Li, Lei. Invisible Image Watermarks Are Provably Removable Using Generative AI. Advances in Neural Information Processing Systems (NeurIPS)

    work page

  3. [3]

    SynSig2Vec: Forgery-Free Learning of Dynamic Signature Representations by Sigma Lognormal-Based Synthesis and 1D CNN

    Lai, Songxuan and Jin, Lianwen and Zhu, Yecheng and Li, Zhe and Lin, Luojun. SynSig2Vec: Forgery-Free Learning of Dynamic Signature Representations by Sigma Lognormal-Based Synthesis and 1D CNN. IEEE Transactions on Pattern Analysis and Machine Intelligence. doi:10.1109/TPAMI.2021.3087619

    arXiv

  4. [4]

    Strategic Change , year =

    McConaghy, Masha and McMullen, Greg and Parry, Glenn and McConaghy, Trent and Holtzman, David , title =. Strategic Change , year =. doi:10.1002/jsc.2146 , publisher =

    work page doi:10.1002/jsc.2146

  5. [5]

    Score Normalization for Text-Independent Speaker Verification Systems

    Auckenthaler, Roland and Carey, Michael and Lloyd-Thomas, Harvey. Score Normalization for Text-Independent Speaker Verification Systems. Digital Signal Processing

    work page

  6. [6]

    Writer Identification Using GMM Supervectors and Exemplar-SVMs

    Christlein, Vincent and Bernecker, David and H \"o nig, Florian and Maier, Andreas and Angelopoulou, Elli. Writer Identification Using GMM Supervectors and Exemplar-SVMs. Pattern Recognition

    work page

  7. [7]

    and Fang, Chen and Jin, Hailin and Hertzmann, Aaron and Collomosse, John and Belongie, Serge

    Wilber, Michael J. and Fang, Chen and Jin, Hailin and Hertzmann, Aaron and Collomosse, John and Belongie, Serge. BAM! The Behance Artistic Media Dataset for Recognition Beyond Photography. Proceedings of the IEEE International Conference on Computer Vision (ICCV)

    work page

  8. [8]

    WikiArt: Visual Art Encyclopedia

    WikiArt. WikiArt: Visual Art Encyclopedia

    work page

Show all 43 references
  1. [9]

    OmniArt: A Large-Scale Artistic Benchmark

    Strezoski, Gjorgji and Worring, Marcel. OmniArt: A Large-Scale Artistic Benchmark. ACM Transactions on Multimedia Computing, Communications, and Applications

  2. [10]

    2016 , howpublished =

    2016

  3. [11]

    and Bruckman, Amy S

    Fiesler, Casey and Feuston, Jessica L. and Bruckman, Amy S. Understanding Copyright Law in Online Creative Communities. Proceedings of the ACM Conference on Computer Supported Cooperative Work & Social Computing (CSCW)

  4. [12]

    C2PA Technical Specification, Version 2.4

    Coalition for Content Provenance and Authenticity. C2PA Technical Specification, Version 2.4

  5. [13]

    Large-Scale Classification of Fine-Art Paintings: Learning the Right Metric on the Right Feature

    Saleh, Babak and Elgammal, Ahmed M. Large-Scale Classification of Fine-Art Paintings: Learning the Right Metric on the Right Feature. International Journal for Digital Art History. doi:10.11588/dah.2016.2.23376

    2016 doi

  6. [14]

    and Tanaka, Kiyoshi

    Tan, Wei Ren and Chan, Chee Seng and Aguirre, Hern \'a n E. and Tanaka, Kiyoshi. Ceci n'est pas une pipe: A Deep Convolutional Network for Fine-Art Paintings Classification. Proceedings of the IEEE International Conference on Image Processing (ICIP)

  7. [15]

    Picasso, Matisse, or a Fake? Automated Analysis of Drawings at the Stroke Level for Attribution and Authentication

    Elgammal, Ahmed and Kang, Yan and Den Leeuw, Milko. Picasso, Matisse, or a Fake? Automated Analysis of Drawings at the Stroke Level for Attribution and Authentication. Proceedings of the AAAI Conference on Artificial Intelligence

  8. [16]

    HMM-Based On-Line Signature Verification: Feature Extraction and Signature Modeling

    Fi \'e rrez-Aguilar, Julian and Ortega-Garcia, Javier and Ramos, Daniel and Gonzalez-Rodriguez, Joaquin. HMM-Based On-Line Signature Verification: Feature Extraction and Signature Modeling. Pattern Recognition Letters

  9. [17]

    Effectiveness of Pen Pressure, Azimuth, and Altitude Features for Online Signature Verification

    Muramatsu, Daigo and Matsumoto, Takashi. Effectiveness of Pen Pressure, Azimuth, and Altitude Features for Online Signature Verification. Proceedings of the International Conference on Biometrics (ICB)

  10. [18]

    DeepSign: Deep On-Line Signature Verification

    Tolosana, Ruben and Vera-Rodriguez, Ruben and Fierrez, Julian and Ortega-Garcia, Javier. DeepSign: Deep On-Line Signature Verification. IEEE Transactions on Biometrics, Behavior, and Identity Science

  11. [19]

    Static-Dynamic Interaction Networks for Offline Signature Verification

    Li, Huan and Wei, Ping and Hu, Ping. Static-Dynamic Interaction Networks for Offline Signature Verification. Proceedings of the AAAI Conference on Artificial Intelligence

  12. [20]

    Offline Signature Verification Using Online Handwriting Registration

    Qiao, Yu and Liu, Jianzhuang and Tang, Xiaoou. Offline Signature Verification Using Online Handwriting Registration. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

  13. [21]

    How Do Humans Sketch Objects?

    Eitz, Mathias and Hays, James and Alexa, Marc. How Do Humans Sketch Objects?. ACM Transactions on Graphics

  14. [22]

    The Sketchy Database: Learning to Retrieve Badly Drawn Bunnies

    Sangkloy, Patsorn and Burnell, Nathan and Ham, Cusuh and Hays, James. The Sketchy Database: Learning to Retrieve Badly Drawn Bunnies. ACM Transactions on Graphics

  15. [23]

    The Quick, Draw! Dataset

    Google Creative Lab. The Quick, Draw! Dataset

  16. [24]

    The DIDI Dataset: Digital Ink Diagram Data

    Gervais, Philippe and Deselaers, Thomas and Aksan, Emre and Hilliges, Otmar. The DIDI Dataset: Digital Ink Diagram Data. arXiv preprint arXiv:2002.09303

    2002 arXiv

  17. [25]

    FS-COCO: Towards Understanding of Freehand Sketches of Common Objects in Context

    Chowdhury, Pinaki Nath and Sain, Aneeshan and Bhunia, Ayan Kumar and Xiang, Tao and Gryaditskaya, Yulia and Song, Yi-Zhe. FS-COCO: Towards Understanding of Freehand Sketches of Common Objects in Context. Proceedings of the European Conference on Computer Vision (ECCV)

  18. [26]

    OpenSketch: A Richly-Annotated Dataset of Product Design Sketches

    Gryaditskaya, Yulia and Sypesteyn, Mark and Hoftijzer, Jan Willem and Pont, Sylvia and Durand, Fr \'e do and Bousseau, Adrien. OpenSketch: A Richly-Annotated Dataset of Product Design Sketches. ACM Transactions on Graphics

  19. [27]

    DifferSketching: How Differently Do People Sketch 3D Objects?

    Xiao, Chufeng and Su, Wanchao and Liao, Jing and Lian, Zhouhui and Song, Yi-Zhe and Fu, Hongbo. DifferSketching: How Differently Do People Sketch 3D Objects?. ACM Transactions on Graphics

  20. [28]

    EMOTHAW: A Novel Database for Emotional State Recognition from Handwriting and Drawing

    Likforman-Sulem, Laurence and Esposito, Anna and Faundez-Zanuy, Marcos and Cl \'e men c on, St \'e phan and Cordasco, Gennaro. EMOTHAW: A Novel Database for Emotional State Recognition from Handwriting and Drawing. IEEE Transactions on Human-Machine Systems

  21. [29]

    A New Learning Paradigm: Learning Using Privileged Information

    Vapnik, Vladimir and Vashist, Akshay. A New Learning Paradigm: Learning Using Privileged Information. Neural Networks

  22. [30]

    Unifying Distillation and Privileged Information

    Lopez-Paz, David and Bottou, L \'e on and Sch \"o lkopf, Bernhard and Vapnik, Vladimir. Unifying Distillation and Privileged Information. Proceedings of the International Conference on Learning Representations (ICLR)

  23. [31]

    Privileged Prior Information Distillation for Image Matting

    Lyu, Cheng and Xie, Jiake and Xu, Bo and Lu, Cheng and Huang, Han and Huang, Xin and Wu, Ming and Zhang, Chuang and Tang, Yong. Privileged Prior Information Distillation for Image Matting. Proceedings of the AAAI Conference on Artificial Intelligence

  24. [32]

    FedCMD: A Federated Cross-Modal Knowledge Distillation for Drivers' Emotion Recognition

    Bano, Saira and Tonellotto, Nicola and Cassar \`a , Pietro and Gotta, Alberto. FedCMD: A Federated Cross-Modal Knowledge Distillation for Drivers' Emotion Recognition. ACM Transactions on Intelligent Systems and Technology

  25. [33]

    and Freeman, William T

    Owens, Andrew and Wu, Jiajun and McDermott, Josh H. and Freeman, William T. and Torralba, Antonio. Ambient Sound Provides Supervision for Visual Learning. Proceedings of the European Conference on Computer Vision (ECCV)

  26. [34]

    Evaluation of Handwriting Kinematics and Pressure for Differential Diagnosis of Parkinson's Disease

    Drot \'a r, Peter and Mekyska, Ji r \'i and Rektorov \'a , Irena and Masarov \'a , Lucia and Sm \'e kal, Zden e k and Faundez-Zanuy, Marcos. Evaluation of Handwriting Kinematics and Pressure for Differential Diagnosis of Parkinson's Disease. Artificial Intelligence in Medicine

  27. [35]

    Apple Pencil Safari API Test

    Ding, Shu. Apple Pencil Safari API Test

  28. [36]

    and Impedovo, Donato and Malik, Muhammad Imran and Pirlo, Giuseppe and Plamondon, R \'e jean

    Diaz, Moises and Ferrer, Miguel A. and Impedovo, Donato and Malik, Muhammad Imran and Pirlo, Giuseppe and Plamondon, R \'e jean. A Perspective Analysis of Handwritten Signature Technology. ACM Computing Surveys

  29. [37]

    Fine-tuning Convolutional Neural Networks for Fine Art Classification

    Cetinic, Eva and Lipic, Tomislav and Grgic, Sonja. Fine-tuning Convolutional Neural Networks for Fine Art Classification. Expert Systems with Applications

  30. [38]

    The Rijksmuseum Challenge: Museum-Centered Visual Recognition

    Mensink, Thomas and van Gemert, Jan. The Rijksmuseum Challenge: Museum-Centered Visual Recognition. Proceedings of the ACM International Conference on Multimedia Retrieval (ICMR)

  31. [39]

    arXiv preprint arXiv:1703.07737 , year =

    Alexander Hermans and Lucas Beyer and Bastian Leibe , title =. arXiv preprint arXiv:1703.07737 , year =

    arXiv

  32. [40]

    Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology (UIST '25) , year =

    Tan, Sean Rui Xiang and Chan, Mun Choon and Han, Jun , title =. Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology (UIST '25) , year =

  33. [41]

    30th USENIX Security Symposium (USENIX Security 21) , year =

    Ramesh, Soundarya and Xiao, Rui and Maiti, Anindya and Lee, Jong Taek and Ramprasad, Harini and Kumar, Ananda and Jadliwala, Murtuza and Han, Jun , title =. 30th USENIX Security Symposium (USENIX Security 21) , year =

  34. [42]

    2018 IEEE Symposium on Security and Privacy (SP) , year =

    Han, Jun and Chung, Albert Jin and Sinha, Manal Kumar and Harishankar, Madhumitha and Pan, Shijia and Noh, Hae Young and Zhang, Pei and Tague, Patrick , title =. 2018 IEEE Symposium on Security and Privacy (SP) , year =

    2018

  35. [43]

    Proceedings of the 23rd Annual International Conference on Mobile Systems, Applications and Services (MobiSys '25) , year =

    Sun, Bangjie and Chan, Mun Choon and Han, Jun , title =. Proceedings of the 23rd Annual International Conference on Mobile Systems, Applications and Services (MobiSys '25) , year =

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.