Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Bridging the Sim2Real Gap: Vision Encoder Pre-Training for Visuomotor Policy Transfer

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper proposes two offline scores—Domain Invariance Score and Action Score—that it claims predict whether a pre-trained vision encoder will help a simulated robot policy transfer to reality.

desk verdict A useful offline benchmark of 23 encoders with two interpretable scores, but the headline 'predictors of Sim2Real transfer' is unsupported because no policy was ever deployed. read the letter →

arxiv 2501.16389 v2 pith:JVUOZLR3 submitted 2025-01-26 cs.RO cs.CV

Yash Yardi , Samuel Biruduganti , Lars Ankile This is my paper · ORCID
classification cs.ROcs.CV
keywords Sim2RealvisionencodervisuomotorpolicytransferlearningroboticmanipulationdomaininvarianceActionScore
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 tries to establish that the Sim2Real gap for vision-based robot policies can be predicted offline by two cheap scores computed from frozen vision-encoder embeddings, before any policy is trained or any robot is run. The Domain Invariance Score (DIS) measures how similarly an encoder represents simulated and real images, while the Action Score (AS) measures how much task-relevant action information a linear readout can recover from the embeddings. Testing 23 pre-trained encoders on a simulated and real table-construction dataset, the paper finds that manipulation-pretrained encoders score higher on AS, convolutional networks score higher on DIS than vision transformers, and the best encoder (MCR) combines both. If the scores are valid predictors, researchers could rank encoder backbones for a new robot task using only pre-collected image-action data.

What carries the argument

The load-bearing machinery is the pair of offline metrics. DIS is computed by standardizing embeddings, projecting them with PCA to a fixed dimension, applying per-feature min-max normalization, and taking one minus the dimension-normalized Euclidean distance between the simulated and real centroids; it is designed to quantify domain invariance. AS is computed by training a linear probe to predict ground-truth robot actions from frozen embeddings and taking one minus the normalized validation mean-squared error; it quantifies how much action-relevant information the representation retains. Grad-CAM saliency maps are used only as a supporting qualitative check, to confirm that high-scoring encoders attend to the robot's end-effectors and objects in both domains. These two scores together are what the paper proposes as an offline proxy for Sim2Real transferability.

What would settle it

Take the two encoders at the extremes of the paper's DIS-AS plot—MCR and DinoV2-B—train identical visuomotor policies on top of their frozen embeddings in simulation, deploy both policies on the same physical robot for the same table-construction task, and compare success rates; the paper's claim predicts a clear ordering matching the offline scores.

Watch

Extended reading notes

Core claim

The paper's central claim is that DIS and AS are complementary predictors of Sim2Real transferability, and that the joint two-axis ranking identifies which pre-trained vision encoders should be used to learn visuomotor policies that transfer from simulation to reality. On the paper's evidence, MCR—a self-supervised CNN pre-trained on large-scale robot manipulation data—is the best encoder because it sits near the top-left of the DIS–AS plot: its simulated and real embeddings align closely, and actions are linearly recoverable from them. The authors also report systematic patterns: manipulation-pretrained encoders such as R3M and HRP consistently beat ImageNet-pretrained baselines on AS, CNNs show stronger domain invariance than ViTs, and parameter count does not predict either score. The paper stops short of showing that these offline scores actually order real-robot policy success, stating that such validation should come from real-world deployments.

Load-bearing premise

The scores computed offline from embeddings correctly predict how well a policy trained on those embeddings will transfer to a physical robot, even though the paper does not run any such policy on a real robot.

Editorial extensions

If this is right

  • Practitioners can rank candidate vision encoders for a manipulation task using only labeled sim/real image pairs plus recorded actions, without running any policy rollouts.
  • The two scores can be combined into a single 'Sim2Real score' and packaged as a reusable library or benchmark, as the paper proposes in its future-work section.
  • Model-selection practice should shift from parameter count to pre-training data relevance: manipulation-pretrained encoders should be preferred over larger general-purpose encoders for robot control.
  • For tasks with a strong domain shift, CNN backbones should be favored over vision transformers, since CNNs in this study achieve higher DIS.
  • If the scores are valid, they enable open-access leaderboards comparing encoders across tasks and domains without the cost of physical robot evaluations.

Reading between the lines

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

  • If the offline scores are validated on real robots, the framework could also be used to monitor drift over time: re-computing DIS on newly collected real images would flag when an encoder's sim-real alignment degrades.
  • Because DIS is a centroid-distance measure, it may not capture local structure; encoders with identical centroids could still differ in how they order nearby sim and real samples, so the score could under-reward representations that are useful for fine-grained manipulation.
  • The task-dependence of the ranking is untested; if the framework is applied to other tasks, the rank order of encoders may shift, meaning the safest use is to re-run the framework per task rather than treat any single encoder as universally best.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes an offline framework for evaluating 23 pre-trained vision encoders on their suitability for sim-to-real visuomotor transfer. Two metrics are introduced: the Domain Invariance Score (DIS), defined as the inverse dimension-normalized Euclidean distance between the centroids of simulated and real embedding sets after PCA projection, and the Action Score (AS), defined as the accuracy of a linear probe that predicts robot actions from frozen embeddings. The authors report that manipulation-pretrained encoders achieve higher AS, CNN-based encoders achieve higher DIS than ViTs, and that MCR is the best overall encoder because it combines high DIS with high AS. From these results they conclude that DIS and AS are complementary predictors of Sim2Real transferability. Section V.A explicitly states that the framework does not include direct validation on physical robots and that the predictive value should ultimately be confirmed in real-world policy deployments.

Significance. If validated, the framework would fill a genuine need: inexpensive, offline selection of pre-trained visual encoders before expensive robot rollouts. The paper's strengths are the breadth of the encoder suite (23 models), the explicit algorithmic definitions of DIS and AS, and the transparent acknowledgment of the missing deployment validation. The metric definitions are external to the conclusions, so the framework is not circular in a derivation sense. However, the headline claim that DIS and AS predict sim-to-real transfer is not supported by any downstream policy experiment; the paper currently offers a descriptive benchmark of embedding statistics, not a validated predictor. The significance therefore depends on future validation rather than on evidence presented in this manuscript.

major comments (3)
  1. [Abstract, Section IV.A, Section V.A] The central claim that DIS and AS are 'complementary predictors of Sim2Real transferability' is not empirically tested. The paper trains no visuomotor policy on any of the 23 encoders, deploys no policy in a real environment, and reports no rank correlation between (DIS, AS) and transfer success. Section V.A explicitly concedes that the framework 'does not include direct validation on physical robots' and that the predictive value 'should ultimately be confirmed in real-world policy deployments.' This is load-bearing because the abstract and Section IV.A interpret MCR as 'the highest-performing encoder' on the strength of its position in Fig. 4. As written, the results support a descriptive claim about embedding statistics, not a predictive claim about sim-to-real transfer. Please either add a transfer experiment (even a simulated distractor or lighting shift with a trained policy) or weaken the title, abstract, and conclusions to describe an offline benchmark whose predictive value remains to be established.
  2. [Eq. (1), Section III.C] The DIS definition rewards collapsed representations: a constant embedding yields DIS=1, and the score contains no term that penalizes loss of task-relevant information. Consequently the 'complementarity' of DIS and AS is partly built into the metric pair rather than being an independent empirical discovery. The paper does not provide a diagnostic showing that encoders with high DIS preserve the information needed for control, apart from the joint use of AS. Without a transfer experiment, the reader cannot tell whether high DIS is selecting for invariance or for triviality. Please add a control or a quantitative statement of what DIS alone does and does not measure.
  3. [Section III.C, Section III.D, Fig. 4] The quantitative results are reported as point estimates without error bars or repeated trials, and key hyperparameters are missing. The PCA target dimension d* is never specified, the probe's training epochs, batch size, learning rate, and action normalization are not reported, and the two metrics are computed on a single split of the 50,000-frame dataset. Because the headline patterns (CNN-based encoders stronger on DIS, manipulation-pretrained encoders stronger on AS) are based on differences in Fig. 4, these omissions make it difficult to assess whether the observed differences are reliable. Please report d*, probe hyperparameters, split sizes, and variability across seeds or bootstrap resamples.
minor comments (5)
  1. [Section III.D, Eq. (2)] The Action Score subtracts normalized validation MSE from 1, but no reference baseline (such as predicting the mean action) is provided; reporting improvement over a mean predictor would make the score more interpretable.
  2. [Section III.B] The text says all embeddings are normalized, but the details (L2 normalization, standardization, or per-feature min-max scaling) are only fully specified later for DIS; please state which normalization is applied before the linear probe for AS.
  3. [Section IV.A] The claim that parameter count 'did not correlate with improved performance' is based on visual inspection of Fig. 4; a rank correlation coefficient would support this statement.
  4. [Fig. 4] The figure would be easier to read with a legend for marker size and with error bars or confidence regions for each encoder.
  5. [References] Reference [29] is not the standard archival source for vision transformers; please cite the original ViT paper or a peer-reviewed equivalent.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DIS and AS are operationally defined metrics, and the transfer-prediction claim is explicitly left for future real-world validation rather than derived from the metrics.

full rationale

The paper's metrics are not derived from its conclusions. DIS (Eq. 1) is defined directly as a centroid-distance statistic on PCA-projected embeddings, and AS (Eq. 2) as the validation accuracy of a linear action probe; neither metric is fit to the paper's headline claim. The finding that MCR scores high on both axes is a measurement result, and the statement that 'the best-performing models combine both properties' is an interpretation of the scatter plot, not an equation reducing the conclusion to the metric definitions. The only place the metrics are called 'predictors of Sim2Real transferability' is an extrapolation that the authors themselves flag in Section V.A: the framework 'does not include direct validation on physical robots' and predictive value 'should ultimately be confirmed in real-world policy deployments.' That is an unvalidated external-validity claim, not a circular derivation. The sole self-citation (ref [13], which includes co-author Ankile) appears in the dataset paragraph, but the dataset is attributed to the external FurnitureBench benchmark [12], so the self-citation is not load-bearing. No self-citation chain, imported uniqueness theorem, or ansatz-via-citation is present. Accordingly, no circular step can be exhibited with a specific reduction.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The framework depends on two proxy assumptions about what offline scores mean, one joint PCA design choice, and several unreported hyperparameters. No new entities are introduced. The axioms are not trivial: if linear probing does not reflect policy-relevant features, or if centroid distance does not capture the sim-real shift that hurts policies, the benchmark rankings lose their stated meaning.

free parameters (3)
  • PCA target dimension d* = not reported
    DIS in Eq. 1 normalizes centroid distance by sqrt(d*), but d* is never specified; changing it changes the scale of the score and the ordering.
  • linear probe hyperparameters (epochs, batch size, learning rate) = not reported
    AS in Eq. 2 depends on training a probe; hyperparameters are not given, so scores may vary with training configuration.
  • action normalization and preprocessing = not reported
    AS is defined over raw action MSE; without normalization details, a model can score high simply by predicting low-variance actions or by exploiting action scale.
assumptions (3)
  • domain assumption Linear probing of frozen embeddings reliably measures task-relevant information for visuomotor control.
    AS in Eq. 2 equates action recoverability with representation quality; introduced in Section III.D and used for all rankings.
  • domain assumption Mean embedding distance captures the Sim2Real shift relevant to policy failure.
    DIS in Eq. 1 assumes the domain gap is summarized by the centroid gap in a PCA-projected space, forgoing higher-order distribution statistics; Section III.C.
  • ad hoc to paper Projecting all embeddings through a jointly fitted PCA into dimension d* gives a fair common comparison space.
    Algorithm 1 fits PCA on the union of sim and real embeddings before computing DIS; no separate fit or sensitivity analysis over d* is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging the Sim2Real Gap: Vision Encoder Pre-Training for Visuomotor Policy Transfer." pith.science (2026). https://pith.science/paper/JVUOZLR3

@misc{pith2026250116389,
  author       = {Pith},
  title        = {Pith review of: Bridging the Sim2Real Gap: Vision Encoder Pre-Training for Visuomotor Policy Transfer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JVUOZLR3}},
  note         = {Machine review of arXiv:2501.16389}
}
read the original abstract

Simulation offers a scalable and efficient alternative to real-world data collection for learning visuomotor robotic policies. However, the simulation-to-reality, or Sim2Real distribution shift -- introduced by employing simulation-trained policies in real-world environments -- frequently prevents successful policy transfer. We present an offline framework to evaluate the performance of using large-scale pre-trained vision encoders to address the Sim2Real gap. We examine a diverse collection of encoders, assessing their ability to extract features necessary for robot control (Action Score) while remaining invariant to task-irrelevant environmental variations (Domain Invariance Score). Evaluating 23 encoders, we reveal patterns across architectures, pre-training datasets, and parameter scales. Our findings show that manipulation-pretrained encoders consistently achieve higher Action Scores, CNN-based encoders demonstrate stronger domain invariance than ViTs, and the best-performing models combine both properties, underscoring DIS and AS as complementary predictors of Sim2Real transferability.

Figures

Figures reproduced from arXiv: 2501.16389 by the authors.

Figure 1
Figure 1. Three Random Frames from the Dataset used. Top Row is Simulation Images; Bottom Row is Real Images. To the human eye, the simulated and real scenes are visually highly similar. Still, machine learning systems struggle. Rdx which is supposed to be a low-dimensional, information￾dense representation of the contents of the image. Twenty-three encoders are analyzed across all metrics to evaluate the performance of vario… view at source ↗
Figure 2
Figure 2. Grad-CAM Maps Generated With DinoV2-B [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. shows that MCR (8 in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Plot Of Each Encoder’s Performance In Domain Invariance and Action Inference, where point size represents the number of parameters, point color represents CNN/ViT-based, and point shape represents manipulation/general pre-training. See Table I for corresponding encoder…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Foundation Model Driven Robotics: A Comprehensive Review

    cs.RO 2025-07 conditional novelty 2.0 of 10

    A review of foundation-model-driven robotics that synthesizes recent work across perception, planning, control, HRI, simulation, and sim-to-real transfer, and highlights open challenges.

Reference graph

Works this paper leans on

37 extracted references · 10 canonical work pages · cited by 1 Pith paper

  1. [1]

    On autonomous robots,

    G. A. Bekey, “On autonomous robots,” p. 143–146, 1998. [Online]. Available: https://doi.org/10.1017/S0269888998002033

    work page doi:10.1017/s0269888998002033 1998

  2. [2]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, and B. Burchfiel, “Diffusion policy: Visuomotor policy learning via action diffusion,”

    work page

  3. [3]

    Self-supervised correspondence in visuomotor policy learning,

    P. Florence, L. Manuelli, and R. Tedrake, “Self-supervised correspondence in visuomotor policy learning,” pp. 492–499,

    work page

  4. [4]

    Real-world robot learning with masked visual pre-training,

    I. Radosavovic, T. Xiao, S. James, P. Abbeel, J. Malik, and T. Darrell, “Real-world robot learning with masked visual pre-training,” 2022. [Online]. Available: https://arxiv.org/abs/2210.03109

    arXiv 2022

  5. [5]

    Transferring policy of deep reinforcement learning from simulation to reality for robotics,

    H. Ju, R. Juan, R. Gomez, K. Nakamura, and G. Li, “Transferring policy of deep reinforcement learning from simulation to reality for robotics,” pp. 1077–1087, 2022. [Online]. Available: https: //doi.org/10.1038/s42256-022-00573-6

    doi:10.1038/s42256-022-00573-6 2022

  6. [6]

    Renaissance: Investigating the pretraining of vision-language encoders,

    C. K. Clayton Fields, “Renaissance: Investigating the pretraining of vision-language encoders,” 2024. [Online]. Available: https: //arxiv.org/abs/2411.06657

    work page arXiv 2024

  7. [7]

    What do we learn from a large-scale study of pre-trained visual representations in sim and real environments?

    S. Silwal, K. Yadav, T. Wu, J. Vakil, A. Majumdar, and S. Arnaud, “What do we learn from a large-scale study of pre-trained visual representations in sim and real environments?” pp. 17 515–17 521,

    work page

  8. [8]

    What makes pre-trained visual representations successful for robust manipulation?

    K. Burns, Z. Witzel, J. I. Hamid, T. Yu, C. Finn, and K. Hausman, “What makes pre-trained visual representations successful for robust manipulation?” 2024. [Online]. Available: https://openreview.net/forum?id=A1hpY5RNiH

    work page 2024

Show all 37 references
  1. [9]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255

    2009

  2. [10]

    Available: https://doi.org/10.1109/ICRA57147.2024

    [Online]. Available: https://doi.org/10.1109/ICRA57147.2024. 10610218

    2024

  3. [11]

    Light- weight probing of unsupervised representations for reinforcement learning,

    W. Zhang, A. GX-Chen, V . Sobal, Y . LeCun, and N. Carion, “Light- weight probing of unsupervised representations for reinforcement learning,” 2024. [Online]. Available: https://arxiv.org/abs/2208.12345

    2024 arXiv

  4. [12]

    Furniturebench: Reproducible real-world benchmark for long-horizon complex manipulation,

    M. Heo, Y . Lee, D. Lee, and J. J. Lim, “Furniturebench: Reproducible real-world benchmark for long-horizon complex manipulation,” 2023. [Online]. Available: https://arxiv.org/abs/2305.12821

    2023 arXiv

  5. [13]

    Learning to manipulate anywhere: A visual generalizable framework for reinforcement learning,

    Z. Yuan, T. Wei, S. Cheng, G. Zhang, Y . Chen, and H. Xu, “Learning to manipulate anywhere: A visual generalizable framework for reinforcement learning,” 2024. [Online]. Available: https: //openreview.net/forum?id=jart4nhCQr

    2024

  6. [14]

    R3m: A universal visual representation for robot manipulation,

    S. Nair, A. Rajeswaran, V . Kumar, C. Finn, and A. Gupta, “R3m: A universal visual representation for robot manipulation,” 2022. [Online]. Available: https://arxiv.org/abs/2203.12601

    2022 arXiv

  7. [16]

    From imitation to refinement – residual rl for precise assembly,

    L. Ankile, A. Simeonov, I. Shenfeld, M. Torne, and P. Agrawal, “From imitation to refinement – residual rl for precise assembly,”

  8. [17]

    Available: https://arxiv.org/abs/2407.16677

    [Online]. Available: https://arxiv.org/abs/2407.16677

    arXiv

  9. [18]

    Where are we in the search for an artificial visual cortex for embodied intelligence?

    A. Majumdar, K. Yadav, S. Arnaud, Y . J. Ma, C. Chen, and S. Silwal, “Where are we in the search for an artificial visual cortex for embodied intelligence?” 2024. [Online]. Available: https://arxiv.org/abs/2303.18240

    2024 arXiv

  10. [19]

    Searching for mobilenetv3,

    A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, and M. Tan, “Searching for mobilenetv3,” 2019. [Online]. Available: https://arxiv.org/abs/1905.02244

    2019 arXiv

  11. [20]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” 2020. [Online]. Available: https://arxiv.org/abs/1905.11946

    2020 arXiv

  12. [21]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” 2015. [Online]. Available: https://arxiv.org/abs/1409.1556

    2015 arXiv

  13. [22]

    Masked visual pre-training for motor control,

    T. Xiao, I. Radosavovic, T. Darrell, and J. Malik, “Masked visual pre-training for motor control,” 2022. [Online]. Available: https://arxiv.org/abs/2203.06173

    2022 arXiv

  14. [23]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, and Z. Zhang, “Swin transformer: Hierarchical vision transformer using shifted windows,”

  15. [24]

    Robots pre-train robots: Manipulation-centric robotic representation from large-scale robot datasets,

    G. Jiang, Y . Sun, T. Huang, H. Li, Y . Liang, and H. Xu, “Robots pre-train robots: Manipulation-centric robotic representation from large-scale robot datasets,” 2024. [Online]. Available: https: //arxiv.org/abs/2410.22325

    2024 arXiv

  16. [25]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, and T. Unterthiner, “An image is worth 16x16 words: Transformers for image recognition at scale,” 2021. [Online]. Available: https: //arxiv.org/abs/2010.11929

    2021 arXiv

  17. [26]

    Vip: Towards universal visual reward and representation via value-implicit pre-training,

    Y . J. Ma, S. Sodhani, D. Jayaraman, O. Bastani, V . Kumar, and A. Zhang, “Vip: Towards universal visual reward and representation via value-implicit pre-training,” 2023. [Online]. Available: https://arxiv.org/abs/2210.00030

    2023 arXiv

  18. [27]

    Hrp: Human affordances for robotic pre-training,

    M. K. Srirama, S. Dasari, S. Bahl, and A. Gupta, “Hrp: Human affordances for robotic pre-training,” 2024. [Online]. Available: https://arxiv.org/abs/2407.18911

    2024 arXiv

  19. [28]

    An introduction to convolutional neural networks,

    K. O’Shea and R. Nash, “An introduction to convolutional neural networks,” 2015. [Online]. Available: https://arxiv.org/abs/1511.08458

    2015 arXiv

  20. [29]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, and S. Agarwal, “Learning transferable visual models from natural language supervision,” 2021. [Online]. Available: https: //arxiv.org/abs/2103.00020

    2021 arXiv

  21. [30]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, and V . Khalidov, “Dinov2: Learning robust visual features without supervision,” 2024. [Online]. Available: https://arxiv.org/abs/2304. 07193

    2024

  22. [31]

    A survey on contrastive self-supervised learning,

    A. Jaiswal, A. R. Babu, M. Z. Zadeh, D. Banerjee, and F. Makedon, “A survey on contrastive self-supervised learning,” Technologies, vol. 9, no. 1, 2021. [Online]. Available: https: //www.mdpi.com/2227-7080/9/1/2

    2021

  23. [32]

    Compressive-projection principal component analysis,

    J. E. Fowler, “Compressive-projection principal component analysis,” IEEE Transactions on Image Processing, vol. 18, no. 10, pp. 2230–2242, 2009. [Online]. Available: https://doi.org/10.1109/TIP. 2009.2025089

    2009

  24. [33]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” 2015. [Online]. Available: https://arxiv.org/abs/ 1512.03385

    2015 arXiv

  25. [34]

    How do vision transformers work?

    N. Park and S. Kim, “How do vision transformers work?” 2022. [Online]. Available: https://arxiv.org/abs/2202.06709

    2022 arXiv

  26. [35]

    Grad-cam: Visual explanations from deep networks via gradient-based localization,

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-cam: Visual explanations from deep networks via gradient-based localization,”International Journal of Computer Vision, vol. 128, no. 2, p. 336–359, Oct. 2019. [Online]. Available: http://dx.doi.o...

    2019 doi

  27. [2020]

    Available: https://doi.org/10.1109/LRA.2019.2956365

    [Online]. Available: https://doi.org/10.1109/LRA.2019.2956365

    2019

  28. [2021]

    Available: https://arxiv.org/abs/2103.14030

    [Online]. Available: https://arxiv.org/abs/2103.14030

    arXiv

  29. [2024]

    Available: https://arxiv.org/abs/2303.04137

    [Online]. Available: https://arxiv.org/abs/2303.04137

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.