Pith. sign in

REVIEW 4 major objections 5 minor 30 references

For budgeted image classification, routing images by confidence — easy images to cheap decision points, hard images to expensive ones — improves accuracy at a fixed budget compared to random routing.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-07-31 23:18 UTC pith:CQUO4K5W

load-bearing objection Clean formalization of budgeted classification with a useful trade-off analysis; experiments are suggestive but need error bars and a fairer baseline. the 4 major comments →

arxiv 2607.23997 v1 pith:CQUO4K5W submitted 2026-07-27 cs.AI

Exploring Budgeted Image Classification with Content-Sensitive Resource Allocation

Athanasios G. Papadopoulos This is my paper
classification cs.AI MSC 68T4590C1090C05
keywords budgeted image classificationresource allocation integer programcontent-sensitive allocationdecision pointsprocessing treetrade-off ratioconfidence-based routingcomputational budget
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper asks how to maximize classification accuracy when a compute budget is fixed but the environment can change. It frames the task as an integer resource-allocation program that assigns images to decision points of different cost and accuracy; since the program is NP-hard, it relaxes it to a linear program that only decides how many images go to each decision point. The key move is a content-sensitive layer on top: images are sorted by a cheap model's top-1 confidence, easy images go to cheaper decision points, and hard images go to expensive, more accurate ones. Experiments on satellite imagery with two architectures — an image-resolution ensemble and a multi-scale attention model — show this routing beats uniform random allocation at nearly every budget. A trade-off ratio identifies which decision points are worth keeping.

Core claim

The central claim is that this confidence-sorted allocation is a near-optimal approximation to the NP-hard assignment problem. The paper shows that for both tested systems, the content-sensitive strategy yields higher accuracy than the content-agnostic one across nearly the full range of budgets; it also explains why: sorting by top-1 probability at a parent decision point separates images that the cheap model is confident about (and is likely to get right) from images that need the extra capacity of a more expensive decision point. The paper additionally derives a ratio condition — the drop in error per extra unit of cost must beat a competing decision point's — and shows that decision poin

What carries the argument

The processing tree, which arranges decision points by execution cost and dependency, and the sorting heuristic at line 13 of Algorithm 1, which orders images by top-1 softmax probability to estimate difficulty. The trade-off ratio α_{k,r} = (E_r − E_k)/(C_k − C_r) is the paper's theoretical instrument: it predicts, for a given reference decision point r, which more expensive decision point k gives the best error reduction per extra FLOP, and reveals which decision points are never selected.

Load-bearing premise

The entire content-sensitive gain rests on the assumption that an image a cheap model labels with high top-1 confidence is also easy for more expensive decision points; the paper's own failure analysis shows this breaks when decision points specialize to different data modes, and no calibration or validation of this proxy is given.

What would settle it

Sort the test set by top-1 probability at the cheapest decision point and compute the error of a more expensive decision point within each confidence decile; if the error does not decrease monotonically with confidence (or if the lowest-confidence decile is not the most difficult), the sorting heuristic stops paying for itself. A dataset where confidence is miscalibrated, such as adversarially perturbed images, should make content-sensitive allocation fall back to random-level accuracy.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • For any classifier with multiple exit points or model sizes, a budget change requires only re-solving a small linear program and re-sorting a batch; the allocation adapts without retraining.
  • Decision points that do not achieve the maximal trade-off ratio for any reference point can be removed with no loss in optimal accuracy; this gives a simple pruning criterion.
  • The confidence-sorting heuristic generalizes across architecture families: it works both for ensembles of models with different resolution inputs and for a single multi-scale attention model.
  • The failure case when decision points specialize to different data modes means routing should not rely solely on confidence; joint accuracy across decision points can exceed any single one, and the allocation should account for complementarity.

Where Pith is reading between the lines

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

  • The top-1 probability proxy is never validated against true per-image difficulty; a calibrated difficulty score or an auxiliary predictor might extend the gains to cases where confidence is miscalibrated.
  • The framework's dependence on FLOPs as the budget metric is arbitrary; the same linear program and trade-off ratio hold for latency, energy, or monetary cost, provided per-decision-point costs are additive.
  • When decision points specialize to distinct data modes, the paper's observation suggests a testable extension: compute joint accuracy matrices and route images to whichever decision point is most likely to be correct, not merely the most expensive one.
  • A concrete test of the core heuristic on other datasets, such as natural images with class imbalance, would reveal whether confidence-sorting transfers or is specific to satellite imagery.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper addresses budgeted image classification: given a batch of images, a set of decision points (DPs) with different costs/accuracies, and a computational budget, it seeks an assignment of images to DPs that maximizes accuracy. It formulates the problem as an integer program (Eq. 1), relaxes it to a linear program (Eq. 2), and proposes two allocation strategies: a content-agnostic (random) strategy that solves the LP and assigns images randomly according to the resulting proportions, and a content-sensitive strategy (Algorithm 1) that sorts images by the top-1 softmax probability from a cheap DP and routes the supposedly easier images to cheaper DPs and harder images to more expensive DPs. Experiments on fMoW with a multi-resolution EfficientNet ensemble and the TNet hard-attention model are reported, showing that content-sensitive allocation yields higher accuracy than random allocation under most budgets. The paper also derives a trade-off ratio (Eq. 7b) that explains why some DPs (e.g., DP4 of TNet) never receive allocation, and it discusses failure cases when DPs specialize to different data modes (§5.2.4).

Significance. The formal LP formulation and the trade-off-ratio analysis are clean and correct, and the idea of using a cheap DP's confidence to route images across DPs is practically appealing, with the potential to improve budgeted inference without training a gating network. The paper is honest about limitations, including the simplifying assumption E_i^k=E_k and the observed failure when DPs specialize. However, the central empirical claim—that content-sensitive allocation is superior under almost all budgets—rests on an unvalidated confidence proxy and a comparison that confounds content-sensitivity with differences in effective DP costs. If these issues are addressed, the paper would be a useful contribution to resource-efficient classification; as it stands, the evidence is suggestive but not conclusive.

major comments (4)
  1. [§4.3, Algorithm 1 line 13] The entire content-sensitive strategy depends on using the top-1 softmax probability from the current DP as a monotone proxy for image difficulty across all DPs. No calibration, rank-correlation, or transfer analysis is provided to support this proxy. The paper's own §5.2.4 shows a concrete failure: between EN-B0-448 and EN-B0-896, high confidence at DP2 does not imply easy classification at DP3, and the content-sensitive accuracy drops as more images are routed to DP3. Without a validated proxy or a diagnostic for when it holds, the claim that content-sensitive allocation 'leads both TNet and EN-B0 ensemble to higher accuracy under almost all budgets' (end of §5.2.3) is not established as a general property; it may be an artifact of the specific DP pairs and dataset.
  2. [§5.2.3, Figure 6 and Table 1] The comparison between content-agnostic and content-sensitive allocation is confounded. Table 1 shows that the effective cost of DP2 and DP3 is higher in the content-sensitive case (1.93 vs 1.54 GFLOPs for DP2; 8.11 vs 6.18 for DP3) because DPs are executed sequentially rather than in parallel. Consequently, for the same nominal budget B, the LP (Eq. 2) produces different allocation counts n_k under the two strategies. The observed accuracy difference could therefore be partly due to different allocation proportions, not to content-based routing. To isolate the effect of content-sensitivity, a baseline is needed that uses the same sequential processing tree and the same n_k but selects images randomly (i.e., Algorithm 1 without the sorting step). Without this control, the headline claim overstates what the experiments show.
  3. [§5.2.4, §6] The failure case described in §5.2.4—where joint accuracy between DP2 and DP3 is only ~78%, implying partial overlap in which images each DP can classify correctly—directly contradicts the assumption E_i^k=E_k and undermines the generality of the content-sensitive gains. The paper acknowledges this in §6 as a limitation, but it is a load-bearing issue for the central claim. The authors should either restrict the claim to settings where DPs are nested or gradually informative (as with TNet) or provide a way to detect when DPs specialize before applying content-sensitive routing. As written, the conclusion 'content-sensitive heuristics can offer significant accuracy benefits' ( §7) is too broad.
  4. [§5.2.3, Figure 6] The experimental results are reported without any uncertainty quantification. Both strategies include randomness: the initial assignment of images to DPs in Algorithm 1 (lines 7–8) is random, and TNet's location sampling may be stochastic. With a single run, the claim of improvement 'under almost all budgets' cannot be assessed for statistical significance, especially in budget regions where the gap is small. At minimum, the authors should provide error bars or confidence intervals over several random seeds, or bootstrap over test batches.
minor comments (5)
  1. [§4.1] The text states 'Linear program 1 is NP-Hard' but program 1 is the integer program; the LP relaxation (program 2) is polynomial-time solvable. Please correct the wording.
  2. [§5.2.3] Typo: 'random and context-sensitive allocation strategy' should read 'content-sensitive allocation strategy.'
  3. [§5.2.2, Eq. (7)] The argmax in Eq. (7) has no tie-breaking rule. In data such as Table 3, if two trade-off ratios are equal, the optimal LP solution would be indifferent; specifying a tie-break (e.g., lower cost) would make the algorithm deterministic.
  4. [Figure 6] The figure is dense: four curves plus markers for DP boundaries. Consider using distinct line styles and adding a legend directly in the figure; also label the budget axis explicitly (GFLOPs).
  5. [Algorithm 1] Step 13 sorts by 'top-1 probability' but the meaning of 'probability' is not defined (softmax output). State that it is the softmax probability of the predicted class, and note whether it is used as a ranking only—this would clarify that no calibrated probabilities are claimed.

Circularity Check

0 steps flagged

No significant circularity: the central accuracy comparison is a held-out evaluation, and the LP allocation is computed from validation errors and fixed costs rather than from the test results it explains.

full rationale

The paper's central claim—that content-sensitive allocation yields higher accuracy under almost all budgets—is an empirical result obtained by running the pre-trained EN-B0 ensemble and TNet on the held-out fMoW test set. The allocation probabilities n_k are computed by solving the linear program in Eq. 2 using validation-set error rates E_k and fixed FLOP costs C_k (Section 4.2). These quantities are inputs to the experiments, not targets fitted from test accuracy. The content-sensitive sorting in Algorithm 1 line 13 uses the current decision point's top-1 softmax probability as a difficulty proxy; this is a heuristic, and the paper explicitly acknowledges its failure mode in Section 5.2.4 and Section 6 ("we set E_i^k = E_k ... This is a strong assumption, and can lead to suboptimal performance"). That is a validity limitation, not circularity, because the proxy is not constructed from the test accuracies it is used to explain. The trade-off-ratio analysis in Section 5.2.2 follows algebraically from Equations 4 and 6, and the observation that DP4 is never selected is a derived property of the LP, not an assumption equivalent to the result. The self-citation [15] supplies the TNet architecture and trained fMoW models; it is prior published work providing the experimental testbed, and it is not invoked as a load-bearing justification for the current paper's allocation conclusions. The missing sequential-random baseline with identical n_k is a genuine experimental confound, but it does not make any reported quantity equal to an input by construction. Overall, no circular step that reduces the paper's claims to its own inputs was found.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard LP analysis and a strong domain assumption that softmax confidence reflects difficulty. The comparison also lacks a controlled baseline that would isolate the routing effect from the extra pre-pass cost. No new physical or theoretical entities are introduced; processing tree and trade-off ratio are conceptual abstractions.

axioms (5)
  • domain assumption Top-1 softmax probability from a cheap DP is a monotone proxy for image difficulty (higher confidence = easier, likely correct on more expensive DPs).
    Used in Algorithm 1 line 13 to sort images and assign 'harder' (lower confidence) images to more accurate DPs. Not proven; Section 5.2.4 shows a failure case.
  • domain assumption Expected misclassification probability is the same for all images at a given DP (E_i^k = E_k).
    Section 4.2 assumes this to reduce the IP to LP (2); acknowledged as a strong assumption in Limitations.
  • domain assumption Computational cost C_k is independent of image content.
    Stated in Section 4.1; needed for the LP counts to be valid.
  • domain assumption Processing dependencies between DPs can be represented as a tree with parallel nodes at each level.
    Section 4.3 defines processing tree; required for the content-sensitive algorithm; may not hold for all adjustable systems.
  • standard math Standard LP duality / convex-hull optimality.
    Used in Section 5.2.2 to derive trade-off ratios and the condition for DP suitability.

pith-pipeline@v1.3.0-alltime-deepseek · 13629 in / 16191 out tokens · 147140 ms · 2026-07-31T23:18:52.927928+00:00 · methodology

0 comments
read the original abstract

The ever-growing adoption of Artificial Intelligence (AI) creates the need to deploy Deep Neural Networks in a variety of computational environments. We consider dynamic environments, where computational requirements are subject to change, and we pose the following question: How do we adjust the complexity of an AI classification system, in order to maximize its accuracy, while meeting changing computational constraints? We call this problem Budgeted Image Classification, and we formally formulate it as a resource allocation integer program. Given a computational budget, a batch of images, and a classification system that can make decisions with varying complexity (it has multiple decision points), we explore strategies to allocate images to decision points, in order to maximize accuracy within the available budget. The original integer program is NP-Hard, so, we propose a continuous relaxation, leading to a content-agnostic allocation strategy which assigns images to decision points without considering their particular content. We address this issue by proposing a content-sensitive strategy, that we experimentally show it leads to superior performance. We theoretically study the behavior of our strategies, deriving conditions that must be satisfied by decision points to be suitable for budgeted classification. We analyze fails cases, offering insights for future research directions.

Figures

Figures reproduced from arXiv: 2607.23997 by Athanasios G. Papadopoulos.

Figure 1
Figure 1. Figure 1: Three unrolled processing levels of TNet [ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Processing tree of the TNet model (see Section [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Two processing trees of EN-B0 ensemble, which con￾sists of EN-B0-224, EN-B0-448, and EN-B0-896 models (see Section 3.1). The 3 models constitute 3 different decision points. (a) In content-agnostic allocation, DPs are executed in parallel, since images are assigned randomly. (b) In content-sensitive allo￾cation, DPs are executed sequentially, in order of increasing com￾putational cost; first EN-B0-224, the… view at source ↗
Figure 4
Figure 4. Figure 4: Optimal allocation probabilities p = {pk, k = 1, 2, ..., 5} of TNet’s decision points for the processing tree of [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Optimal allocation probabilities p = {pk, k = 1, 2, 3} of EN-B0 ensemble’s decision points for the processing tree of [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Classification accuracy of TNet and EN-B [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

30 extracted references · 14 linked inside Pith

  1. [1]

    Revisiting resnets: Improved training and scaling strategies.arXiv preprint arXiv:2103.07579, 2021

    Irwan Bello, William Fedus, Xianzhi Du, Ekin D Cubuk, Aravind Srinivas, Tsung-Yi Lin, Jonathon Shlens, and Bar- ret Zoph. Revisiting resnets: Improved training and scaling strategies.arXiv preprint arXiv:2103.07579, 2021. 1

    Pith/arXiv arXiv 2021

  2. [2]

    Adaptive neural networks for efficient inference

    Tolga Bolukbasi, Joseph Wang, Ofer Dekel, and Venkatesh Saligrama. Adaptive neural networks for efficient inference. InInternational Conference on Machine Learning, pages 527–536. PMLR, 2017. 2

    2017

  3. [3]

    Lan- guage models are few-shot learners.Advances in neural in- formation processing systems, 33:1877–1901, 2020

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Lan- guage models are few-shot learners.Advances in neural in- formation processing systems, 33:1877–1901, 2020. 1

    1901

  4. [4]

    Functional map of the world

    Gordon Christie, Neil Fendley, James Wilson, and Ryan Mukherjee. Functional map of the world. InCVPR, 2018. 5

    2018

  5. [5]

    Spatially adaptive computation time for residual networks

    Michael Figurnov, Maxwell D Collins, Yukun Zhu, Li Zhang, Jonathan Huang, Dmitry Vetrov, and Ruslan Salakhutdinov. Spatially adaptive computation time for residual networks. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1039– 1048, 2017. 2

    2017

  6. [6]

    Dynamic channel pruning: Feature boosting and suppression.arXiv preprint arXiv:1810.05331,

    Xitong Gao, Yiren Zhao, Łukasz Dudziak, Robert Mullins, and Cheng-zhong Xu. Dynamic channel pruning: Feature boosting and suppression.arXiv preprint arXiv:1810.05331,

    Pith/arXiv arXiv

  7. [7]

    Accurate, large mini- batch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017

    Priya Goyal, Piotr Doll ´ar, Ross Girshick, Pieter Noord- huis, Lukasz Wesolowski, Aapo Kyrola, Andrew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large mini- batch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017. 1

    Pith/arXiv arXiv 2017

  8. [8]

    Adaptive computation time for recurrent neural networks.arXiv preprint arXiv:1603.08983, 2016

    Alex Graves. Adaptive computation time for recurrent neural networks.arXiv preprint arXiv:1603.08983, 2016. 2

    Pith/arXiv arXiv 2016

  9. [9]

    Searching for mo- bilenetv3

    Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mo- bilenetv3. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 1314–1324, 2019. 1

    2019

  10. [10]

    Mobilenets: Efficient convolu- tional neural networks for mobile vision applications.arXiv preprint arXiv:1704.04861, 2017

    Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco An- dreetto, and Hartwig Adam. Mobilenets: Efficient convolu- tional neural networks for mobile vision applications.arXiv preprint arXiv:1704.04861, 2017. 1

    Pith/arXiv arXiv 2017

  11. [11]

    Multi-scale dense networks for resource efficient image classification.arXiv preprint arXiv:1703.09844, 2017

    Gao Huang, Danlu Chen, Tianhong Li, Felix Wu, Laurens van der Maaten, and Kilian Q Weinberger. Multi-scale dense networks for resource efficient image classification.arXiv preprint arXiv:1703.09844, 2017. 2

    Pith/arXiv arXiv 2017

  12. [12]

    Improved techniques for training adaptive deep net- works

    Hao Li, Hong Zhang, Xiaojuan Qi, Ruigang Yang, and Gao Huang. Improved techniques for training adaptive deep net- works. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 1891–1900, 2019. 2

    1900

  13. [13]

    Dynamic computational time for visual attention

    Zhichao Li, Yi Yang, Xiao Liu, Feng Zhou, Shilei Wen, and Wei Xu. Dynamic computational time for visual attention. In Proceedings of the IEEE International Conference on Com- puter Vision Workshops, pages 1199–1209, 2017. 2

    2017

  14. [14]

    Changing model behavior at test-time using reinforcement learning.arXiv preprint arXiv:1702.07780, 2017

    Augustus Odena, Dieterich Lawson, and Christopher Olah. Changing model behavior at test-time using reinforcement learning.arXiv preprint arXiv:1702.07780, 2017. 2

    Pith/arXiv arXiv 2017

  15. [15]

    Hard-attention for scalable image classification.Advances in Neural Information Processing Systems, 34:14694–14707,

    Athanasios Papadopoulos, Pawel Korus, and Nasir Memon. Hard-attention for scalable image classification.Advances in Neural Information Processing Systems, 34:14694–14707,

  16. [16]

    Adaptative inference cost with convolutional neural mixture models

    Adria Ruiz and Jakob Verbeek. Adaptative inference cost with convolutional neural mixture models. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 1872–1881, 2019. 2

    2019

  17. [17]

    Outra- geously large neural networks: The sparsely-gated mixture- of-experts layer.arXiv preprint arXiv:1701.06538, 2017

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outra- geously large neural networks: The sparsely-gated mixture- of-experts layer.arXiv preprint arXiv:1701.06538, 2017. 2

    Pith/arXiv arXiv 2017

  18. [18]

    Energy and policy considerations for deep learning in nlp

    Emma Strubell, Ananya Ganesh, and Andrew McCallum. Energy and policy considerations for deep learning in nlp. arXiv preprint arXiv:1906.02243, 2019. 1

    Pith/arXiv arXiv 1906

  19. [19]

    Efficientnet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. InInternational Conference on Machine Learning, pages 6105–6114. PMLR,

  20. [20]

    Efficientnetv2: Smaller mod- els and faster training.arXiv preprint arXiv:2104.00298,

    Mingxing Tan and Quoc V Le. Efficientnetv2: Smaller mod- els and faster training.arXiv preprint arXiv:2104.00298,

    Pith/arXiv arXiv

  21. [21]

    Branchynet: Fast inference via early exiting from deep neural networks

    Surat Teerapittayanon, Bradley McDanel, and Hsiang-Tsung Kung. Branchynet: Fast inference via early exiting from deep neural networks. In2016 23rd International Con- ference on Pattern Recognition (ICPR), pages 2464–2469. IEEE, 2016. 2

    2016

  22. [22]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. InInternational Conference on Machine Learning, pages 10347–10357. PMLR, 2021. 1

    2021

  23. [23]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. 1

    Pith/arXiv arXiv 2023

  24. [24]

    Convolutional networks with adaptive inference graphs

    Andreas Veit and Serge Belongie. Convolutional networks with adaptive inference graphs. InProceedings of the Euro- pean Conference on Computer Vision (ECCV), pages 3–18,

  25. [25]

    Dynamic convolu- tions: Exploiting spatial sparsity for faster inference

    Thomas Verelst and Tinne Tuytelaars. Dynamic convolu- tions: Exploiting spatial sparsity for faster inference. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2320–2329, 2020. 2

    2020

  26. [26]

    Skipnet: Learning dynamic routing in convolutional networks

    Xin Wang, Fisher Yu, Zi-Yi Dou, Trevor Darrell, and Joseph E Gonzalez. Skipnet: Learning dynamic routing in convolutional networks. InProceedings of the European Conference on Computer Vision (ECCV), pages 409–424,

  27. [27]

    Glance and focus: a dynamic approach to reducing spatial redundancy in image classification.arXiv preprint arXiv:2010.05300, 2020

    Yulin Wang, Kangchen Lv, Rui Huang, Shiji Song, Le Yang, and Gao Huang. Glance and focus: a dynamic approach to reducing spatial redundancy in image classification.arXiv preprint arXiv:2010.05300, 2020. 2

    Pith/arXiv arXiv 2010

  28. [28]

    Not all images are worth 16x16 words: Dynamic vision transformers with adaptive sequence length.arXiv preprint arXiv:2105.15075, 2021

    Yulin Wang, Rui Huang, Shiji Song, Zeyi Huang, and Gao Huang. Not all images are worth 16x16 words: Dynamic vision transformers with adaptive sequence length.arXiv preprint arXiv:2105.15075, 2021. 2

    Pith/arXiv arXiv 2021

  29. [29]

    Resolution adaptive networks for efficient infer- ence

    Le Yang, Yizeng Han, Xi Chen, Shiji Song, Jifeng Dai, and Gao Huang. Resolution adaptive networks for efficient infer- ence. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 2369–2378,

  30. [30]

    Slimmable neural networks.arXiv preprint arXiv:1812.08928, 2018

    Jiahui Yu, Linjie Yang, Ning Xu, Jianchao Yang, and Thomas Huang. Slimmable neural networks.arXiv preprint arXiv:1812.08928, 2018. 2

    Pith/arXiv arXiv 2018