Pith. sign in

REVIEW 4 major objections 4 minor 50 references

Frequency Composition for Compressed and Domain-Adaptive Neural Networks

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 CoDA, a frequency-composition training and test-time adaptation pipeline, lets heavily quantized networks match or beat full-precision networks under domain shift.

desk verdict A strong empirical combo of LFC training and frequency-aware BN adaptation for quantized models, but the headline gains rely on test-set-tuned radius and the weakest baseline. read the letter →

arxiv 2505.20890 v1 pith:T4DTWTIU submitted 2025-05-27 cs.CV cs.AI

Yoojin Kwon , Hongjun Suh , Wooseok Lee , Taesik Gong , Songyi Han , Hyung-Sin Kim This is my paper
classification cs.CVcs.AI
keywords quantization-awaretrainingtest-timeadaptationfrequencydecompositiondomainshiftbatchnormalizationmodelcompressionlow-frequencycomponentscorruptionrobustness
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

The paper's claim is that compression and test-time adaptation should be solved as one problem, and that frequency content is the key that ties them together. It proposes CoDA, which restricts quantization-aware training to low-frequency image components and then adapts the compact model at test time with a frequency-aware batch-norm scheme that treats low- and high-frequency features differently. The payoff is empirical: heavily quantized models become more accurate under domain shift than larger full-precision models with standard adaptation. A sympathetic reader would take away the practical prospect of smaller on-device models that are not only cheaper but also more reliable when the deployment environment drifts.

What carries the argument

The mechanism is frequency decomposition: a 2D Fourier transform with radial low-pass and high-pass filters splits each image into $x_{\mathrm{lfc}}$ and $x_{\mathrm{hfc}}$ with $x = x_{\mathrm{lfc}} + x_{\mathrm{hfc}}$. CoDA trains with LFC-only images during quantization-aware training (LFC QAT), and at test time applies Frequency-Aware Batch Normalization (FABN), which band-pass filters each BN layer's input feature, evolves low-frequency statistics from the source running mean and variance via an exponential moving average, takes high-frequency statistics directly from the current batch, and adds the two. The additive recombination of means and variances is the step that carries the argument.

What would settle it

Compute, on a corrupted test set, the covariance between the low- and high-frequency bands of the BN input features at each layer. If that cross-covariance is large relative to the individual variances, the additive rule $\hat{\sigma}_t^2 = \hat{\sigma}_{\mathrm{lfc},t}^2 + \hat{\sigma}_{\mathrm{hfc},t}^2$ drops real signal and FABN should be miscalibrated. A second check: run ordinary prediction-time batch-statistics adaptation without the band split on the same LFC-trained model; if it matches FABN's accuracy, the frequency split is not the active ingredient.

Watch

Extended reading notes

Core claim

The central discovery is that a quantized model trained only on low-frequency image components learns a flatter, more domain-invariant representation, and that at test time adapting a band-split batch normalization, with source-initialized EMA statistics for low frequencies and current-batch statistics for high frequencies, lets a 2-bit model surpass full-precision TTA baselines by up to 7.96 percentage points on CIFAR10-C and 5.37 percentage points on ImageNet-C while using 4 to 16 times fewer bits. The paper presents this as a unified pipeline that wraps existing QAT and TTA methods rather than replacing them.

Load-bearing premise

The load-bearing premise is that a low/high-frequency split that separates general from domain-specific information in input images does the same inside hidden-layer activations, and that adding the two bands' means and variances loses nothing important.

Editorial extensions

If this is right

  • A 2-bit quantized model trained and adapted with CoDA can beat full-precision TTA baselines on CIFAR10-C and ImageNet-C, by up to 7.96 and 5.37 percentage points, while using 4-16x fewer bits.
  • CoDA is additive rather than competitive: it raises the accuracy of existing QAT methods such as LSQ and LQ, and of TTA methods such as NORM, TENT, and SAR, when combined with any of them.
  • The improvement transfers to lightweight architectures such as MobileNet-v3 and EfficientNet-b0, which also surpass their full-precision TTA counterparts.
  • Because the low-frequency statistics remain anchored to the source distribution, CoDA holds up at small test batch sizes where NORM, TENT, and SAR degrade sharply.
  • Even under continually changing corruptions, CoDA is competitive with CoTTA, a method explicitly designed for continual domain shift.

Reading between the lines

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

  • The method's core split could be stress-tested layer by layer: if low- and high-frequency activations are strongly correlated at deeper layers, the additive statistics in Eq. (5) would need cross-covariance corrections, and the advantage should shrink there.
  • One testable extension is to apply the frequency split to the affine-parameter branch of TTA, giving low- and high-frequency channels independent adaptation rates; the paper's compatibility results suggest this but do not explore it.
  • LFC QAT's flatter loss landscape hints that low-frequency filtering might serve as a cheap regularizer for quantized training generally, potentially complementing sharpness-based penalties without extra gradient computation.
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 proposes CoDA, a frequency-composition framework that jointly addresses model compression and test-time domain adaptation. During training, CoDA applies quantization-aware training (QAT) to low-frequency components (LFC) of images; at test time, it adapts batch-normalization statistics through Frequency-Aware Batch Normalization (FABN), which filters intermediate activations into LFC and HFC, updates LFC statistics with source-initialized EMA, uses current-batch HFC statistics, and combines them by summing means and variances. The method is evaluated on CIFAR10-C, ImageNet-C, ImageNet-R, and ImageNet-Sketch with LSQ and LQ quantizers at 2/4/8 bits across ResNet, MobileNet, and EfficientNet, and is combined with NORM, TENT, and SAR. The headline claims are accuracy improvements of 7.96pp on CIFAR10-C and 5.37pp on ImageNet-C over full-precision TTA baselines, with 4-16x model-size reduction.

Significance. If fully substantiated, CoDA would be an practically important demonstration that heavily quantized models can be more robust under domain shift than larger full-precision models with standard TTA. The paper's strengths include extensive experiments, per-corruption results with standard deviations for CIFAR10-C and ImageNet-C, integration with multiple TTA baselines, multiple architectures and bitwidths, model-size calculations, and ablations of the two main components. However, the central quantitative claims currently rest on a frequency radius selected using the same corrupted test distributions used for evaluation, and the specific benefit of FABN over standard TTA on the same LFC-trained model is not isolated. These issues must be resolved before the significance of the results can be assessed.

major comments (4)
  1. [Supplementary Section C and Table 2] The frequency radius r is selected using target-domain accuracy on the corrupted test sets. Table 2 reports CIFAR10-C and ImageNet-C accuracy as a function of r, and Section C states that r=8 (CIFAR10) and r=56 (ImageNet) were chosen from this analysis. Since r controls both LFC QAT training (Sec. 3.1-3.2) and FABN test-time filtering (Sec. 3.3), the headline gains of 7.96pp and 5.37pp over full-precision TTA may be inflated by test-set leakage. Please select r on source-domain validation data (or a clean split), and report a sensitivity analysis of the full CoDA pipeline across a range of r values on both datasets.
  2. [Tables 4, 5, and 9] The paper does not compare FABN against standard TTA applied to the same LFC-trained model. In Tables 4 and 5, the TTA baselines (NORM, TENT, SAR) are applied to FFC-trained models, while CoDA rows use LFC-trained models with FABN; the FABN-only row in Table 9 is based on an FFC-trained model with FABN. This does not establish that FABN is better than simply running NORM, TENT, or SAR on an LFC-QAT model. Please add experiments with each standard TTA method applied to the same LFC-trained model and compare against FABN.
  3. [Equation (5), Section 3.3] Equation (5) sets σ̂²_t = σ̂²_lfc,t + σ̂²_hfc,t, which assumes that the cross-covariance between the low-frequency and high-frequency feature activations is zero. No feature-space verification is provided. Please measure the correlation between f_lfc and f_hfc in representative BN layers, or use the exact variance of the sum, and report whether the results change.
  4. [Equation (3), Section 3.3] The EMA coefficient α in Equation (3) is a key hyperparameter of FABN, but its value is never reported anywhere in the main text or supplementary material, and no sensitivity analysis is given. Please report α and show how performance varies with it.
minor comments (4)
  1. [Throughout] There are several typos that should be corrected, including 'informastion' in Section 3.2.1, 'adpated' in the Table 3 caption, 'should should' in Section 3, and 'Domian' in the heading of Supplementary Section G.
  2. [Tables 13 and 14] The last column header 'A VG' appears to be a typo for 'Avg'; please correct it.
  3. [Figure 2] Figure 2 is dense and the connection between the frequency-domain filtering and the BN statistics update is hard to follow; labeling the equations and the train/test phases more explicitly would help.
  4. [Section 3.3] The terms 'FFC' and 'full-frequency' are used interchangeably; please define the abbreviation at first use to avoid ambiguity.

Circularity Check

1 steps flagged · score 2.0 of 10

Mild circularity-adjacent issue: radius r is selected using the target corrupted benchmarks, slightly inflating headline CoDA gains; no equation-level circularity.

  1. fitted input called prediction [Supplementary Section C; Section 3.2.2 Table 2; Section 3.3 (FABN radius); Abstract and Tables 4/5 (headline claims)]
    "Finally, to ensure that LPF preserves a more robust and lower frequency range, we set r=8 for CIFAR10 and r=56 for ImageNet. ... FABN applies two bandpass filters, using the same radius r as the training phase. ... it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA baseline."

    The radius r is a method-defining hyperparameter: it sets the low/high-frequency split for LFC QAT during training and is reused by FABN at test time. It is selected with Table 2, whose columns report accuracy on CIFAR10-C and ImageNet-C, the exact target benchmarks used for the headline evaluation. The paper then reports accuracy gains of 7.96%p and 5.37%p on those same benchmarks, so the target-domain test distribution is used both to choose a central component of the method and to evaluate the method.

full rationale

The core derivation is empirical and largely self-contained: LFC QAT and frequency-aware BN are new combinations of established QAT and TTA building blocks, and the main comparisons are against external baselines. There is no load-bearing self-citation chain or imported uniqueness theorem, and Equation 5's additive combination of LFC/HFC statistics is an unverified modeling assumption rather than a circular derivation, since the paper does not claim it follows from Equation 1 by construction. The only circularity-adjacent step is the choice of radius r from accuracy on the corrupted test benchmarks, which mildly contaminates the headline improvements; this is evaluation leakage and hyperparameter selection on the target distribution, not an equation-level reduction. I therefore score it 2 rather than higher.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central method relies on the empirical assumption that low-frequency content carries domain-invariant information and that the frequency split transfers to intermediate activations. The only explicit free parameters are the filter radius r (selected on the target benchmarks) and the unreported EMA coefficient alpha. No new physical entities are introduced.

free parameters (2)
  • training/test frequency radius r = 8 (CIFAR10), 56 (ImageNet)
    Selected based on robustness accuracy on CIFAR10-C/ImageNet-C in Table 2 and App. C; corresponds to 25% of input size. This is a hand-tuned scalar that controls how much high-frequency information is discarded during training and used at test time.
  • EMA coefficient alpha = not specified in paper
    Used in Eq. 3 for LFC BN statistic EMA; no value or sensitivity analysis is reported in the main text or appendix.
assumptions (4)
  • domain assumption Low-frequency image components are more domain-invariant than high-frequency components.
    Supported by Figure 3 (cosine distances) and prior work [43], but it is the premise for LFC QAT.
  • domain assumption Training on low-pass-filtered images retains enough label-relevant information for classification.
    Depends on radius r; Table 1 shows clean accuracy drops of 1-2% for ImageNet at r=56, which is accepted as a trade-off.
  • ad hoc to paper The variance of the sum of LFC and HFC features equals the sum of their variances (cross-covariance is negligible).
    Eq. 5 adds sigma^2_lfc and sigma^2_hfc; no justification or empirical check of the covariance term is provided.
  • domain assumption Batch normalization statistics of the source model for LFC features remain representative of target LFC features and can be adapted via EMA.
    Core to FABN's LFC stream; Figure 5 provides some evidence but only on quantized models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Frequency Composition for Compressed and Domain-Adaptive Neural Networks." pith.science (2026). https://pith.science/paper/T4DTWTIU

@misc{pith2026250520890,
  author       = {Pith},
  title        = {Pith review of: Frequency Composition for Compressed and Domain-Adaptive Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4DTWTIU}},
  note         = {Machine review of arXiv:2505.20890}
}
read the original abstract

Modern on-device neural network applications must operate under resource constraints while adapting to unpredictable domain shifts. However, this combined challenge-model compression and domain adaptation-remains largely unaddressed, as prior work has tackled each issue in isolation: compressed networks prioritize efficiency within a fixed domain, whereas large, capable models focus on handling domain shifts. In this work, we propose CoDA, a frequency composition-based framework that unifies compression and domain adaptation. During training, CoDA employs quantization-aware training (QAT) with low-frequency components, enabling a compressed model to selectively learn robust, generalizable features. At test time, it refines the compact model in a source-free manner (i.e., test-time adaptation, TTA), leveraging the full-frequency information from incoming data to adapt to target domains while treating high-frequency components as domain-specific cues. LFC are aligned with the trained distribution, while HFC unique to the target distribution are solely utilized for batch normalization. CoDA can be integrated synergistically into existing QAT and TTA methods. CoDA is evaluated on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA baseline.

Figures

Figures reproduced from arXiv: 2505.20890 by the authors.

Figure 1
Figure 1. Effectiveness of CoDA when applied to various models (ResNet18 and ResNet50), TTA methods (NORM [28, 31] and TENT [34]) and QAT method (LSQ [10]) using three bitwidths (2, 4, and 8 bits). We train on ImageNet and evaluate on ImageNet-C. to incoming, unlabeled test data in a source-free manner, a process known as test-time adaptation (TTA) [4, 13, 27– 29, 31, 34, 47]. To meet this requirement, our test-phase procedur… view at source ↗
Figure 2
Figure 2. An illustration of the proposed CoDA. Left: Using Fast Fourier Transformation, an image can be decomposed into HFC consisting of fast-changing patterns (i.e. edges or stripes) and LFC consisting of slow-changing patterns (i.e. smooth shape). During training, CoDA focus on learning generalizable features from LFC rather than irregular patterns in HFC (LFC QAT). Right: At test time, under domain shift, CoDA utilizes f… view at source ↗
Figure 3
Figure 3. Inter-domain distance matrices of LFC and HFC fre [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The loss landscapes of quantized ResNet26 on CIFAR10 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: SSE comparison between LSQ [10] with and without CoDA. SSE is measured between µˆs and µˆlfc,t for the LFC-trained model (CoDA), and between µˆs and µˆt for the FFC-trained model. SSE is gathered over all layers. Quantization level is 2-bit. Here, µlfc,t and σ 2 lfc,t …
Figure 6
Figure 6. Figure 6: Classification accuracy (%) under the effect of different [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: t-SNE visualization of embeddings from FFC/LFC [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: The loss landscapes of full precision ResNet26 on CI [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages

  1. [1]

    R2snet: Scalable domain adaptation for object detection in cloud– based robotic ecosystems via proposal refinement

    Michele Antonazzi, Matteo Luperto, N Alberto Borghese, and Nicola Basilico. R2snet: Scalable domain adaptation for object detection in cloud– based robotic ecosystems via proposal refinement. In 2024 IEEE/RSJ International Conference on Intelli- gent Robots and Systems (IROS), pages 2676–2682. IEEE, 2024. 1

    work page 2024

  2. [2]

    Qgen: On the ability to generalize in quantization aware training

    MohammadHossein AskariHemmat, Ahmadreza Jeddi, Reyhane Askari Hemmat, Ivan Lazarevich, Alexander Hoffman, Sudhakar Sah, Ehsan Saboori, Yvon Savaria, and Jean-Pierre David. Qgen: On the ability to generalize in quantization aware training. arXiv preprint arXiv:2404.11769, 2024. 2

    arXiv 2024

  3. [3]

    Michele Boldo, Mirco De Marchi, Enrico Martini, Stefano Aldegheri, and Nicola Bombieri. Domain- adaptive online active learning for real-time intelligent video analytics on edge devices.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 43(11):4105–4116, 2024. 1

    work page 2024

  4. [4]

    Parameter-free online test-time adaptation

    Malik Boudiaf, Romain Mueller, Ismail Ben Ayed, and Luca Bertinetto. Parameter-free online test-time adaptation. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 8344–8353, 2022. 2

    work page 2022

  5. [5]

    Prentice-Hall, Inc., 1988

    E Oran Brigham.The fast Fourier transform and its applications. Prentice-Hall, Inc., 1988. 1

    work page 1988

  6. [6]

    Pasta: Proportional amplitude spectrum training augmentation for syn-to- real domain generalization

    Prithvijit Chattopadhyay, Kartik Sarangmath, Vivek Vijaykumar, and Judy Hoffman. Pasta: Proportional amplitude spectrum training augmentation for syn-to- real domain generalization. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 19288–19300, 2023. 1, 2

    work page 2023

  7. [7]

    Amplitude-phase recombina- tion: Rethinking robustness of convolutional neural networks in frequency domain

    Guangyao Chen, Peixi Peng, Li Ma, Jia Li, Lin Du, and Yonghong Tian. Amplitude-phase recombina- tion: Rethinking robustness of convolutional neural networks in frequency domain. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 458–467, 2021. 1, 2

    work page 2021

  8. [8]

    Binaryconnect: Training deep neural networks with binary weights during propagations

    Matthieu Courbariaux, Yoshua Bengio, and Jean- Pierre David. Binaryconnect: Training deep neural networks with binary weights during propagations. Advances in neural information processing systems, 28, 2015. 2

    work page 2015

Show all 50 references
  1. [9]

    Imagenet: A large-scale hierarchi- cal image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In2009 IEEE conference on com- puter vision and pattern recognition, pages 248–255. Ieee, 2009. 2, 3, 4, 12

    2009

  2. [10]

    Learned step size quantization.arXiv preprint arXiv:1902.08153, 2019

    Steven K Esser, Jeffrey L McKinstry, Deepika Bablani, Rathinakumar Appuswamy, and Dharmen- dra S Modha. Learned step size quantization.arXiv preprint arXiv:1902.08153, 2019. 2, 3, 4, 5, 6, 8, 12

    1902 arXiv

  3. [11]

    Sharpness-aware minimiza- tion for efficiently improving generalization.arXiv preprint arXiv:2010.01412, 2020

    Pierre Foret, Ariel Kleiner, Hossein Mobahi, and Behnam Neyshabur. Sharpness-aware minimiza- tion for efficiently improving generalization.arXiv preprint arXiv:2010.01412, 2020. 4, 13

    2010 arXiv

  4. [12]

    A survey of quantization methods for efficient neural network inference

    Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. A survey of quantization methods for efficient neural network inference. InLow-Power Computer Vision, pages 291–326. Chapman and Hall/CRC, 2022. 1

    2022

  5. [13]

    Note: Robust continual test-time adaptation against temporal corre- lation.Advances in Neural Information Processing Systems, 35:27253–27266, 2022

    Taesik Gong, Jongheon Jeong, Taewon Kim, Yewon Kim, Jinwoo Shin, and Sung-Ju Lee. Note: Robust continual test-time adaptation against temporal corre- lation.Advances in Neural Information Processing Systems, 35:27253–27266, 2022. 2

    2022

  6. [14]

    Sotta: Robust test-time adaptation on noisy data streams.Advances in Neural Information Processing Systems, 36, 2024

    Taesik Gong, Yewon Kim, Taeckyung Lee, Sorn Chot- tananurak, and Sung-Ju Lee. Sotta: Robust test-time adaptation on noisy data streams.Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  7. [15]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 2, 3, 12

    2016

  8. [16]

    Identity mappings in deep residual networks

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In Computer Vision–ECCV 2016: 14th European Con- ference, Amsterdam, The Netherlands, October 11– 14, 2016, Proceedings, Part IV 14, pages 630–645. Springer, 2016. 3, 6, 12

    2016

  9. [17]

    Benchmark- ing neural network robustness to common corruptions and perturbations.arXiv preprint arXiv:1903.12261,

    Dan Hendrycks and Thomas Dietterich. Benchmark- ing neural network robustness to common corruptions and perturbations.arXiv preprint arXiv:1903.12261,

    1903 arXiv

  10. [18]

    The many faces of robustness: A critical analysis of out- of-distribution generalization

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kadavath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, et al. The many faces of robustness: A critical analysis of out- of-distribution generalization. InProceedings of the IEEE/CVF international c...

    2021

  11. [19]

    Searching for mobilenetv3

    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 mobilenetv3. InProceedings of the IEEE/CVF international conference on computer vi- sion, pages 1314–1324, 2019. 6, 7

    2019

  12. [20]

    Daformer: Improving network architectures and train- ing strategies for domain-adaptive semantic segmen- tation

    Lukas Hoyer, Dengxin Dai, and Luc Van Gool. Daformer: Improving network architectures and train- ing strategies for domain-adaptive semantic segmen- tation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9924–9935, 2022. 1 9

    2022

  13. [21]

    Fsdr: Frequency space domain randomiza- tion for domain generalization

    Jiaxing Huang, Dayan Guan, Aoran Xiao, and Shi- jian Lu. Fsdr: Frequency space domain randomiza- tion for domain generalization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 6891–6902, 2021. 1, 2

    2021

  14. [22]

    Domain adaptable fine-tune distillation framework for advancing farm surveillance.arXiv preprint arXiv:2402.07059, 2024

    Raza Imam, Muhammad Huzaifa, Nabil Man- sour, Shaher Bano Mirza, and Fouad Lamghari. Domain adaptable fine-tune distillation framework for advancing farm surveillance.arXiv preprint arXiv:2402.07059, 2024. 1

    2024 arXiv

  15. [23]

    Qt- dog: Quantization-aware training for domain general- ization.arXiv preprint arXiv:2410.06020, 2024

    Saqib Javed, Hieu Le, and Mathieu Salzmann. Qt- dog: Quantization-aware training for domain general- ization.arXiv preprint arXiv:2410.06020, 2024. 2

    2024 arXiv

  16. [24]

    Neural network quantization with scale- adjusted training

    Qing Jin, Linjie Yang, Zhenyu Liao, and Xiaon- ing Qian. Neural network quantization with scale- adjusted training. InBMVC, 2020. 2

    2020

  17. [25]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 3, 4, 6, 12

    2009

  18. [26]

    Visualizing the loss landscape of neural nets.Advances in neural information process- ing systems, 31, 2018

    Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. Visualizing the loss landscape of neural nets.Advances in neural information process- ing systems, 31, 2018. 4, 5, 13

    2018

  19. [27]

    The norm must go on: Dynamic unsupervised domain adaptation by normalization

    M Jehanzeb Mirza, Jakub Micorek, Horst Possegger, and Horst Bischof. The norm must go on: Dynamic unsupervised domain adaptation by normalization. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 14765– 14775, 2022. 2, 5

    2022

  20. [28]

    Evaluating prediction-time batch normal- ization for robustness under covariate shift.arXiv preprint arXiv:2006.10963, 2020

    Zachary Nado, Shreyas Padhy, D Sculley, Alexan- der D’Amour, Balaji Lakshminarayanan, and Jasper Snoek. Evaluating prediction-time batch normal- ization for robustness under covariate shift.arXiv preprint arXiv:2006.10963, 2020. 2, 6, 7, 13

    2006 arXiv

  21. [29]

    Efficient test-time model adaptation without forget- ting

    Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Yaofo Chen, Shijian Zheng, Peilin Zhao, and Mingkui Tan. Efficient test-time model adaptation without forget- ting. InInternational conference on machine learning, pages 16888–16905. PMLR, 2022. 2

    2022

  22. [30]

    Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023

    Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Zhiquan Wen, Yaofo Chen, Peilin Zhao, and Mingkui Tan. Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023. 2, 5, 6, 7, 13

    2023 arXiv

  23. [31]

    Improving robustness against common corruptions by covariate shift adaptation.Advances in neural infor- mation processing systems, 33:11539–11551, 2020

    Steffen Schneider, Evgenia Rusak, Luisa Eck, Oliver Bringmann, Wieland Brendel, and Matthias Bethge. Improving robustness against common corruptions by covariate shift adaptation.Advances in neural infor- mation processing systems, 33:11539–11551, 2020. 2, 5, 6, 13

    2020

  24. [32]

    Efficientnet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, pages 6105–6114. PMLR, 2019. 6, 7

    2019

  25. [33]

    Visu- alizing data using t-sne.Journal of machine learning research, 9(11), 2008

    Laurens Van der Maaten and Geoffrey Hinton. Visu- alizing data using t-sne.Journal of machine learning research, 9(11), 2008. 8

    2008

  26. [34]

    Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726, 2020

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726, 2020. 2, 5, 6, 7, 13

    2006 arXiv

  27. [35]

    Learning robust global representations by penalizing local predictive power.Advances in neural information processing systems, 32, 2019

    Haohan Wang, Songwei Ge, Zachary Lipton, and Eric P Xing. Learning robust global representations by penalizing local predictive power.Advances in neural information processing systems, 32, 2019. 6, 12

    2019

  28. [36]

    High-frequency component helps explain the generalization of convolutional neural networks

    Haohan Wang, Xindi Wu, Zeyi Huang, and Eric P Xing. High-frequency component helps explain the generalization of convolutional neural networks. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 8684– 8694, 2020. 2, 3

    2020

  29. [37]

    Generalizing to unseen do- mains: A survey on domain generalization.IEEE transactions on knowledge and data engineering, 35 (8):8052–8072, 2022

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and S Yu Philip. Generalizing to unseen do- mains: A survey on domain generalization.IEEE transactions on knowledge and data engineering, 35 (8):8052–8072, 2022. 1

    2022

  30. [38]

    Continual test-time domain adaptation

    Qin Wang, Olga Fink, Luc Van Gool, and Dengxin Dai. Continual test-time domain adaptation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7201–7211,

  31. [39]

    Efficienttrain: Exploring generalized curriculum learning for training visual backbones

    Yulin Wang, Yang Yue, Rui Lu, Tianjiao Liu, Zhao Zhong, Shiji Song, and Gao Huang. Efficienttrain: Exploring generalized curriculum learning for training visual backbones. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 5852–5864, 2023. 2, 3

    2023

  32. [40]

    Learning in the fre- quency domain

    Kai Xu, Minghai Qin, Fei Sun, Yuhao Wang, Yen- Kuang Chen, and Fengbo Ren. Learning in the fre- quency domain. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 1740–1749, 2020. 3

    2020

  33. [41]

    A fourier-based framework for do- main generalization

    Qinwei Xu, Ruipeng Zhang, Ya Zhang, Yanfeng Wang, and Qi Tian. A fourier-based framework for do- main generalization. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 14383–14392, 2021. 1, 2

    2021

  34. [42]

    Fda: Fourier do- main adaptation for semantic segmentation

    Yanchao Yang and Stefano Soatto. Fda: Fourier do- main adaptation for semantic segmentation. InPro- ceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 4085–4095, 2020. 1, 2

    2020

  35. [43]

    A fourier per- 10 spective on model robustness in computer vision.Ad- vances in Neural Information Processing Systems, 32,

    Dong Yin, Raphael Gontijo Lopes, Jon Shlens, Ekin Dogus Cubuk, and Justin Gilmer. A fourier per- 10 spective on model robustness in computer vision.Ad- vances in Neural Information Processing Systems, 32,

  36. [44]

    Adapt-net: A unified ob- ject detection framework for mobile augmented real- ity.IEEE Access, 2024

    Xiangyun Zeng, Siok Yee Tan, and Moham- mad Faidzul Nasrudin. Adapt-net: A unified ob- ject detection framework for mobile augmented real- ity.IEEE Access, 2024. 1

    2024

  37. [45]

    Lq-nets: Learned quantization for highly accurate and compact deep neural networks

    Dongqing Zhang, Jiaolong Yang, Dongqiangzi Ye, and Gang Hua. Lq-nets: Learned quantization for highly accurate and compact deep neural networks. In Proceedings of the European conference on computer vision (ECCV), pages 365–382, 2018. 2, 3, 4, 6, 7, 8

    2018

  38. [46]

    Why quantization improves generalization: Ntk of binary weight neural networks.arXiv preprint arXiv:2206.05916, 2022

    Kaiqi Zhang, Ming Yin, and Yu-Xiang Wang. Why quantization improves generalization: Ntk of binary weight neural networks.arXiv preprint arXiv:2206.05916, 2022. 2

    2022 arXiv

  39. [47]

    Memo: Test time robustness via adaptation and aug- mentation.Advances in neural information processing systems, 35:38629–38642, 2022

    Marvin Zhang, Sergey Levine, and Chelsea Finn. Memo: Test time robustness via adaptation and aug- mentation.Advances in neural information processing systems, 35:38629–38642, 2022. 2

    2022

  40. [48]

    A review of single-source deep unsupervised visual domain adaptation.IEEE Trans- actions on Neural Networks and Learning Systems, 33 (2):473–493, 2020

    Sicheng Zhao, Xiangyu Yue, Shanghang Zhang, Bo Li, Han Zhao, Bichen Wu, Ravi Krishna, Joseph E Gonzalez, Alberto L Sangiovanni-Vincentelli, San- jit A Seshia, et al. A review of single-source deep unsupervised visual domain adaptation.IEEE Trans- actions on Neural Networks and...

    2020

  41. [49]

    Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients.arXiv preprint arXiv:1606.06160,

    Shuchang Zhou, Yuxin Wu, Zekun Ni, Xinyu Zhou, He Wen, and Yuheng Zou. Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients.arXiv preprint arXiv:1606.06160,

    arXiv

  42. [2016]

    Experimental Details We use pre-activation [16] based ResNet [15] models

    2 11 Frequency Composition for Compressed and Domain-Adaptive Neural Networks (Supplementary Material) A. Experimental Details We use pre-activation [16] based ResNet [15] models. All models are trained from scratch. ResNet26 [16] is used for training on CIFAR10 [25], while Re...

Pith tools

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