Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation

T0 review · 4 major / 5 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 The paper's central claim is that the Fisher Information Matrix, approximated as a diagonal-plus-low-rank matrix and built from the KL-divergence gradient, is a better quantization-reconstruction loss than squared-gradient Hessian…

desk verdict Strong empirical PTQ paper with an overclaimed theoretical centerpiece; the DPLR-FIM loss works in practice, but Theorem 3.2 is an approximation stated as exact and a key hyperparameter is missing. read the letter →

arxiv 2506.11543 v1 pith:FC2SAX5N submitted 2025-06-13 cs.CV cs.AIcs.LG

Zhuguanyu Wu , Shihe Wang , Jiayi Zhang , Jiaxin Chen , Yunhong Wang This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords post-trainingquantizationvisiontransformersFisherinformationmatrixKLdivergenceHessianapproximationdiagonalpluslow-ranklow-bitblock-wisereconstruction
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

Post-training quantization of vision transformers collapses at 3-bit precision when the reconstruction loss is built from a Hessian whose diagonal is approximated by squared gradients; the paper argues this surrogate is wrong because Fisher Information is linearly proportional to the KL-divergence gradient, not to its square. It establishes the exact identity $L_{\mathrm{KL}}(\Delta z^{(b)}) = \tfrac{1}{2}\Delta z^{(b)\top} F(z^{(b)}) \Delta z^{(b)}$ and exploits it to design a diagonal-plus-low-rank FIM loss (DPLR-FIM) that keeps both individual output sensitivities and inter-token correlations. Plugging this loss into the standard QDrop block-reconstruction pipeline, with only a plain uniform quantizer, the method reports average gains of 5.31% over the second-best approach at W3/A3 on ImageNet, and lifts accuracy over the MSE baseline by up to 23.04% on ViT-S. The claim matters because it suggests the reconstruction loss, not quantizer flexibility, is the main bottleneck for low-bit ViT deployment.

What carries the argument

The load-bearing object is the DPLR-FIM loss $L_{\mathrm{DPLR}} = \alpha L_{\mathrm{rank}-k} + (1-\alpha) L_{\mathrm{diag}}$, a diagonal-plus-low-rank surrogate for the Fisher Information Matrix. The identity that carries the argument is Theorem 3.2, $L_{\mathrm{KL}}(\Delta z^{(b)}) = \tfrac{1}{2}\Delta z^{(b)\top} F(z^{(b)}) \Delta z^{(b)}$, which makes the FIM linearly recoverable from the KL-divergence gradient via $\nabla L_{\mathrm{KL}} = F \Delta z^{(b)}$. The diagonal term is built from the per-element ratio of gradient to perturbation; the rank-$k$ term is built from $\nabla L_{\mathrm{KL}}$ times the Moore-Penrose inverse of the perturbation matrix, capturing inter-token correlations at $O(ak)$ cost; a progressive strategy increases $k$ during reconstruction.

What would settle it

Compute, for a block-wise quantized ViT at W3/A3, the exact KL divergence $D_{\mathrm{KL}}(p(y;z^{(b)})\,\|\,p(y;z^{(b)}+\Delta z^{(b)}))$ by sampling over the calibration set, and compare it to the DPLR-FIM quadratic form $\tfrac{1}{2}\Delta z^{(b)\top} F_{\mathrm{DPLR}} \Delta z^{(b)}$. If the ratio of the two quantities deviates substantially from 1 across blocks, or if the higher-order remainder of the Taylor expansion is not negligible, the surrogate is not faithful and the reported gains cannot be attributed to the FIM mechanism.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the prevailing Hessian-guided quantization loss in post-training quantization has been using the wrong approximation: replacing the Fisher Information Matrix with squared gradients discards both the variance of the score and all off-diagonal structure. The paper proves, via a second-order Taylor expansion of the KL divergence under the regularity condition, that the quantization loss equals $\frac{1}{2}\Delta z^\top F(z)\Delta z$, and therefore the gradient of the KL loss is $F(z)\Delta z$; FIM is thus recovered from the ratio of the KL gradient to the perturbation, linearly, not as a squared gradient. From that relation it builds three progressively richer surrogates — diagonal, rank-one/low-rank, and their diagonal-plus-low-rank combination — and shows empirically that the DPLR-FIM loss, integrated into QDrop-style block reconstruction with a uniform quantizer, pushes 3-bit and 4-bit accuracy above methods that rely on specialized quantizers. The paper reports consistent gains across ViT, DeiT, and Swin architectures on ImageNet and COCO, with the largest margins at 3-bit quantization.

Load-bearing premise

The paper's central identity is obtained by a second-order Taylor expansion of the KL divergence, so it holds only when the quantization perturbation $\Delta z^{(b)}$ is small enough that higher-order terms can be dropped; the paper does not directly verify this small-perturbation condition at aggressive 3-bit quantization, where the reported gains are largest.

Editorial extensions

If this is right

  • At W3/A3 on ImageNet, FIMA-Q reports 64.09% top-1 on ViT-S and 69.13% on DeiT-S with a plain uniform quantizer, surpassing the second-best optimization-based method by 5.31% on average across architectures.
  • Swapping the squared-gradient diagonal Hessian for the DPLR-FIM loss improves top-1 accuracy over the MSE baseline by 23.04% (ViT-S) and 18.18% (DeiT-S) at W3/A3, with an 8.74% average improvement.
  • Because only a standard uniform quantizer is needed, the accuracy gains do not depend on hardware-unfriendly specialized quantizers such as log or twin-uniform quantizers.
  • In object detection and instance segmentation on COCO, using RPN classification outputs as the task signal, the method achieves the best W4/A4 results in most Mask R-CNN and Cascade Mask R-CNN settings with Swin backbones.
  • Reconstruction time grows linearly with rank $k$ but stays under 480 GPU minutes on a single RTX 4090 for the largest tested model, so the method is practical for deployment.

Reading between the lines

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

  • A natural extension the paper leaves implicit: the same DPLR-FIM loss could be applied to post-training quantization of large language models or diffusion models, where block-wise reconstruction with Hessian surrogates is also common.
  • The results suggest that the reconstruction loss, rather than quantizer flexibility, is the main bottleneck for low-bit ViTs; if so, further gains should come from loss design rather than new quantizers.
  • The fixed rank $k=15$ could be made adaptive per block based on the estimated FIM spectrum, trading accuracy for reconstruction time on simple blocks; this is testable with the released code.
  • The supplementary sample-size ablation indicates reconstruction is more calibration-hungry than FIM estimation, so data-efficient reconstruction techniques could widen FIMA-Q's applicability to smaller calibration sets.
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 / 5 minor

Summary. The paper proposes FIMA-Q, a post-training quantization (PTQ) method for Vision Transformers. It replaces the Hessian-guided loss of prior work with a loss based on an approximation of the Fisher Information Matrix (FIM), arguing that FIM is linearly proportional to the gradient of the KL divergence rather than to the squared gradient. The method computes a diagonal approximation, a low-rank approximation, and combines them into a "diagonal plus low-rank" (DPLR) FIM loss. The loss is used in a block-wise reconstruction framework built on QDrop. Experiments on ImageNet classification and COCO detection/segmentation report strong results, especially at 3-bit quantization, while using only a standard uniform quantizer.

Significance. If the central claim is correct, the method is practically significant: it improves low-bit ViT accuracy substantially over prior PTQ methods using a simple uniform quantizer, avoiding specialized quantizers that are harder to implement in hardware. The empirical results in Tables 1 and 2 are extensive and the improvements at W3/A3 are large. The paper also provides code, which aids reproducibility. However, the theoretical foundation for the proposed loss has gaps: Theorem 3.2 is stated as exact but is proven via a truncated Taylor expansion, the combination weight alpha in Eq. (20) is never specified, and the low-rank FIM is not guaranteed to be positive semidefinite. These issues bear directly on whether the loss is a faithful surrogate for the KL divergence, which is the stated explanation for the empirical gains.

major comments (4)
  1. [Sec. 3.2 / Theorem 3.2, Sec. A.3] Theorem 3.2 asserts the exact equality L_KL(Delta z) = 1/2 Delta z^T F(z) Delta z, but the proof in Sec. A.3 is a second-order Taylor expansion in which the remainder is dropped without a bound. The equality therefore holds only up to O(||Delta z||^3) and requires the perturbation Delta z to be sufficiently small. The paper's largest reported gains occur at 3-bit quantization, where this small-perturbation assumption is not verified. Since Eq. (9) is the basis for all subsequent losses (Eqs. 14, 16, 18, 21), the authors should either restate Theorem 3.2 as an approximation with an explicit remainder term, or provide an experiment that measures the omitted remainder or directly compares the proposed loss with the true KL divergence for quantized blocks.
  2. [Sec. 3.3, Eq. (20)] The scalar alpha in the DPLR-FIM combination F_DPLR = alpha F_rank-k + (1-alpha) F_diag is never specified. Eq. (21) defines the final loss as alpha L_rank-k + (1-alpha) L_diag, and the paper later fixes the rank k = 15 but never gives the value of alpha or any ablation over it. Because alpha directly controls the relative contribution of the low-rank and diagonal terms, the final loss is underspecified and the results cannot be reproduced from the text alone. The authors should state the value used and, ideally, report sensitivity to it.
  3. [Sec. 3.3, Eqs. (17)-(19)] The low-rank FIM F = grad L * Delta z^+ is not guaranteed to be symmetric or positive semidefinite. The paper explicitly relaxes the symmetry constraint, but the quadratic form in Eq. (18) is then not a principled FIM or Hessian surrogate: if the symmetric part is indefinite, the loss can be negative and can be decreased without bound when viewed as a function of Delta z. The authors should either enforce symmetry and positive semidefiniteness (e.g., by symmetrization and projection) or explain why the indefinite form still behaves as a useful objective in their optimization, with supporting evidence.
  4. [Sec. 3.3, Eqs. (13)-(19), Algorithm 1] The notation for Delta z^(b) is inconsistent between a vector and a matrix. In Eqs. (9)-(11) Delta z^(b) is an a-dimensional vector, and Eq. (13) writes it as a sum of vectors. But the low-rank construction in Eqs. (17)-(19) and the progressive rank increase described in the text require Delta z^(b) to be an a x k matrix of k linearly independent perturbations, and the Moore-Penrose inverse formula only makes sense in that interpretation. This ambiguity obscures the rank-k loss and should be clarified by defining the matrix explicitly and adjusting Eq. (13) or its interpretation.
minor comments (5)
  1. [Sec. 4.3] The phrase "promoting the accuracy of MSE by 23.04% and 18.18%" should say "by 23.04 and 18.18 percentage points," since it refers to absolute differences in top-1 accuracy (e.g., 64.09 - 41.05 = 23.04).
  2. [Sec. 4.2] The statement that FIMA-Q surpasses the second-best approach by 5.31% on average at 3-bit quantization should specify the averaging procedure and confirm that the difference is measured in percentage points.
  3. [Sec. 4.1 / Table 4] There are minor typos: "diagnal" in Sec. 3.3, "computer" in Sec. 3.3, and "Nvidia RX 4090" in Table 4 should be "RTX 4090."
  4. [Sec. A.3, Eq. (8)] Eq. (8) writes the KL divergence as a sum over x, while the proof in Sec. A.3 uses an integral over R. The notation should be made consistent, since p(y; z) is a distribution and the domain of y is not specified.
  5. [Supplementary, Table A] Table A in the supplementary material is referenced in Sec. B but is not labeled with a number in the main text; adding a reference such as "see Table A in the supplementary" would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FIMA-Q's loss is derived from standard Fisher/KL identities and validated against external accuracy benchmarks; the sole self-citation is a non-load-bearing baseline.

full rationale

The derivation chain starts from a textbook Fisher-information identity (Theorem 3.1) and the KL-FIM relationship (Theorem 3.2). Although Theorem 3.2 is stated as an equality while the appendix proves it via a second-order Taylor expansion, this is a rigor and approximation concern, not circularity: L_KL is defined by the KL divergence and F by the score covariance, so Eq. (9) is not imposed by definition. The DPLR-FIM losses (Eqs. 14-21) are constructed to be consistent with Eq. (10), but their empirical success is measured by top-1 accuracy on ImageNet and COCO against external baselines; no reported accuracy number is a renamed fit parameter. The only reference to prior work by the same group, AdaLog [31], is used as a comparison baseline, not as a load-bearing premise. Hyperparameters (rank k and mixing weight alpha) are tuned, which is standard model selection, and the ablation in Table 3 and Fig. 3 directly varies the proposed components rather than assuming the conclusion. Therefore no circular step is present.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The method's central loss depends on alpha and rank k, both chosen by hand, and alpha's value is absent. The proofs rely on regularity, matching distribution, and a small-perturbation quadratic approximation that is presented as exact. No invented entities are introduced.

free parameters (2)
  • alpha = not reported
    In Eq. (20), the final DPLR-FIM loss is a weighted sum alpha * L_rank-k + (1-alpha) * L_diag; the paper never states the value or tuning procedure for alpha, making the exact method unreproducible.
  • rank k = 15
    Fixed at 15 for all experiments; Fig. 3 shows accuracy varies with rank, so k is a hand-chosen hyperparameter that affects both accuracy and runtime.
assumptions (5)
  • standard math Regularity condition allowing interchange of differentiation and integration
    Used throughout the proofs in Sec. A to exchange derivatives and integrals; standard but not formally verified for neural network likelihoods.
  • domain assumption Model output distribution matches true data distribution
    Theorem 3.1 and Theorem A.2 require the expected score to be zero, which holds exactly only when the model distribution equals the true data distribution; PTQ calibration data is unlabeled and the quantized model is not the true distribution.
  • domain assumption Second-order Taylor expansion of KL divergence is accurate
    Central Eq. (9) is a quadratic approximation of KL divergence, but the paper treats it as an exact theorem; the accuracy depends on quantization perturbations Delta z being small, which is not established at 3-bit.
  • domain assumption Independence of calibration samples for summing gradients
    Eq. (13) writes the total gradient and perturbation as sums over per-sample values, assuming sample independence; reasonable for random calibration data but not formally guaranteed.
  • domain assumption Invertibility of Delta z^T Delta z for the low-rank approximation
    The rank-k FIM via the Moore-Penrose inverse requires k linearly independent perturbations; the paper uses a progressive rank increase to maintain this, but it is not formally guaranteed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation." pith.science (2026). https://pith.science/paper/FC2SAX5N

@misc{pith2026250611543,
  author       = {Pith},
  title        = {Pith review of: FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FC2SAX5N}},
  note         = {Machine review of arXiv:2506.11543}
}
read the original abstract

Post-training quantization (PTQ) has stood out as a cost-effective and promising model compression paradigm in recent years, as it avoids computationally intensive model retraining. Nevertheless, current PTQ methods for Vision Transformers (ViTs) still suffer from significant accuracy degradation, especially under low-bit quantization. To address these shortcomings, we analyze the prevailing Hessian-guided quantization loss, and uncover certain limitations of conventional Hessian approximations. By following the block-wise reconstruction framework, we propose a novel PTQ method for ViTs, dubbed FIMA-Q. Specifically, we firstly establish the connection between KL divergence and FIM, which enables fast computation of the quantization loss during reconstruction. We further propose an efficient FIM approximation method, namely DPLR-FIM, by employing the diagonal plus low-rank principle, and formulate the ultimate quantization loss. Our extensive experiments, conducted across various vision tasks with representative ViT-based architectures on public datasets, demonstrate that our method substantially promotes the accuracy compared to the state-of-the-art approaches, especially in the case of low-bit quantization. The source code is available at https://github.com/ShiheWang/FIMA-Q.

Figures

Figures reproduced from arXiv: 2506.11543 by the authors.

Figure 1
Figure 1. Illustration on the heatmap of FIM for the class token. (a) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Framework overview of the proposed FIMA-Q method. We follow the block-wise quantization pipeline. For each block, we [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Influence of the rank k on the accuracy of DPLR-FIM on ImageNet. method using the diagonal estimation, low-rank estimation as well as their combination, denoted by Diag-FIM, LR-FIM, and DPLR-FIM, respectively. As shown in Tab. 3, despite incorporating second￾order Hessian information, BRECQ-FIM generally performs worse than the conventional MSE loss, due to the inaccu￾rate estimation on FIM. In contrast, based on ou… view at source ↗

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. When W4A4 Breaks Camouflaged Object Detection: Token-Group Dual-Constraint Activation Quantization

    cs.CV 2026-04 unverdicted novelty 7.0 of 10

    Token-group activation scaling with dual constraints on step-to-dispersion and zero-bin mass restores W4A4 COD accuracy from catastrophic collapse to near-FP32 on CFRN and ESCNet.

Reference graph

Works this paper leans on

39 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    Understanding and overcoming the challenges of efficient transformer quantization

    Yelysei Bondarenko, Markus Nagel, and Tijmen Blankevoort. Understanding and overcoming the challenges of efficient transformer quantization. InEMNLP, 2021. 2

    work page 2021

  2. [2]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agar- wal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz L...

    work page 2020

  3. [3]

    Cascade R-CNN: delv- ing into high quality object detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade R-CNN: delv- ing into high quality object detection. InCVPR, pages 6154– 6162, 2018. 6

    work page 2018

  4. [4]

    MMDetection: Open mmlab detection toolbox and benchmark.arXiv preprint arXiv:1906.07155, 2019

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng, Chenchen Zhu, Tian- heng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu, Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang, Chen Change Loy, and Dahua Lin. MMDetection: Open mmlab detection toolbox and...

    arXiv 1906

  5. [5]

    BERT: pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: pre-training of deep bidirectional trans- formers for language understanding. InNAACL-HLT, pages 4171–4186, 2019. 1

    work page 2019

  6. [6]

    Towards accurate post- training quantization for vision transformer

    Yifu Ding, Haotong Qin, Qinghua Yan, Zhenhua Chai, Junjie Liu, Xiaolin Wei, and Xianglong Liu. Towards accurate post- training quantization for vision transformer. InACM MM, pages 5380–5388, 2022. 2, 3, 6, 7

    work page 2022

  7. [7]

    Reg-ptq: Regression-specialized post-training quantization for fully quantized object detector

    Yifu Ding, Weilun Feng, Chuyan Chen, Jinyang Guo, and Xianglong Liu. Reg-ptq: Regression-specialized post-training quantization for fully quantized object detector. InCVPR,

    work page

  8. [8]

    Packqvit: Faster sub-8-bit vision transformers via full and packed quantization on the mobile

    Peiyan Dong, Lei Lu, Chao Wu, Cheng Lyu, Geng Yuan, Hao Tang, and Yanzhi Wang. Packqvit: Faster sub-8-bit vision transformers via full and packed quantization on the mobile. InNeurIPS, 2023. 1, 2

    work page 2023

Show all 39 references
  1. [9]

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

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  2. [10]

    Esser, Jeffrey L

    Steven K. Esser, Jeffrey L. McKinstry, Deepika Bablani, Rathinakumar Appuswamy, and Dharmendra S. Modha. Learned step size quantization. InICLR, 2020. 1, 2

    2020

  3. [11]

    On the mathematical foundations of theoretical statistics.Philosophical Transactions of the Royal Society of London

    Fisher and Ronald Aylmer. On the mathematical foundations of theoretical statistics.Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character, 222:309–368, 1922. 11

    1922

  4. [12]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InCVPR, pages 770–778, 2016. 1

    2016

  5. [13]

    Girshick

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross B. Girshick. Mask R-CNN. InICCV, pages 2980–2988, 2017. 6

    2017

  6. [14]

    Quantizing deep convolutional networks for efficient inference: A whitepaper.arXiv preprint arXiv:1806.08342, 2018

    Raghuraman Krishnamoorthi. Quantizing deep convolutional networks for efficient inference: A whitepaper.arXiv preprint arXiv:1806.08342, 2018. 1

    2018 arXiv

  7. [15]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Im- agenet classification with deep convolutional neural networks. InNeurIPS, pages 1106–1114, 2012. 1

    2012

  8. [16]

    BRECQ: pushing the limit of post-training quantization by block reconstruction

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. BRECQ: pushing the limit of post-training quantization by block reconstruction. InICLR, 2021. 2, 3, 6, 7, 8

    2021

  9. [17]

    Q-vit: Accurate and fully quantized low-bit vision transformer

    Yanjing Li, Sheng Xu, Baochang Zhang, Xianbin Cao, Peng Gao, and Guodong Guo. Q-vit: Accurate and fully quantized low-bit vision transformer. InNeurIPS, 2022. 1, 2

    2022

  10. [18]

    I-vit: Integer-only quantization for efficient vision transformer inference

    Zhikai Li and Qingyi Gu. I-vit: Integer-only quantization for efficient vision transformer inference. InICCV, pages 17019–17029, 2023. 2

    2023

  11. [19]

    Repq- vit: Scale reparameterization for post-training quantization of vision transformers

    Zhikai Li, Junrui Xiao, Lianwei Yang, and Qingyi Gu. Repq- vit: Scale reparameterization for post-training quantization of vision transformers. InICCV, pages 17227–17236, 2023. 2, 6, 7

    2023

  12. [20]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C. Lawrence Zitnick. Microsoft COCO: common objects in context. In ECCV, pages 740–755, 2014. 6

    2014

  13. [21]

    Fq-vit: Post-training quantization for fully quantized vision transformer

    Yang Lin, Tianyu Zhang, Peiqin Sun, Zheng Li, and Shuchang Zhou. Fq-vit: Post-training quantization for fully quantized vision transformer. InIJCAI, pages 1173–1179, 2022. 2

    2022

  14. [22]

    Noisyquant: Noisy bias-enhanced post-training activation quantization for vision transformers

    Yijiang Liu, Huanrui Yang, Zhen Dong, Kurt Keutzer, Li Du, and Shanghang Zhang. Noisyquant: Noisy bias-enhanced post-training activation quantization for vision transformers. InCVPR, pages 20321–20330, 2023. 6

    2023

  15. [23]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In ICCV, pages 9992–10002, 2021. 6

    2021

  16. [24]

    Outlier-aware slicing for post-training quantization in vision transformer

    Yuexiao Ma, Huixia Li, Xiawu Zheng, Feng Ling, Xuefeng Xiao, Rui Wang, Shilei Wen, Fei Chao, and Rongrong Ji. Outlier-aware slicing for post-training quantization in vision transformer. InICML, 2024. 6

    2024

  17. [25]

    Instance-aware group quantization for vision transform- ers

    Jaehyeon Moon, Dohyung Kim, Junyong Cheon, and Bumsub Ham. Instance-aware group quantization for vision transform- ers. InCVPR, 2024. 2, 6

    2024

  18. [26]

    Up or down? adaptive rounding for post-training quantization

    Markus Nagel, Rana Ali Amjad, Mart van Baalen, Christos Louizos, and Tijmen Blankevoort. Up or down? adaptive rounding for post-training quantization. InICML, pages 7197– 7206, 2020. 2, 4

    2020

  19. [27]

    Bernstein, Alexander C

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael S. Bernstein, Alexander C. Berg, and Li Fei-Fei. Imagenet large scale visual recognition challenge. IJCV, 115(3):211–252, 2015. 6

    2015

  20. [28]

    Very deep convolu- tional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convolu- tional networks for large-scale image recognition. InICLR,

  21. [29]

    Training data-efficient image transformers & distillation through atten- tion

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv´e J´egou. Training data-efficient image transformers & distillation through atten- tion. InICML, pages 10347–10357, 2021. 6

    2021

  22. [30]

    Qdrop: Randomly dropping quantization for extremely low-bit post-training quantization

    Xiuying Wei, Ruihao Gong, Yuhang Li, Xianglong Liu, and Fengwei Yu. Qdrop: Randomly dropping quantization for extremely low-bit post-training quantization. InICLR, 2022. 2, 3, 6, 7, 8

    2022

  23. [31]

    Adalog: Post-training quantization for vision transformers with adaptive logarithm quantizer

    Zhuguanyu Wu, Jiaxin Chen, Hanwen Zhong, Di Huang, and Yunhong Wang. Adalog: Post-training quantization for vision transformers with adaptive logarithm quantizer. InECCV,

  24. [32]

    Dopq- vit: Towards distribution-friendly and outlier-aware post- training quantization for vision transformers.arXiv preprint arXiv:2408.03291, 2024

    Lianwei Yang, Haisong Gong, and Qingyi Gu. Dopq- vit: Towards distribution-friendly and outlier-aware post- training quantization for vision transformers.arXiv preprint arXiv:2408.03291, 2024. 6, 7

    2024 arXiv

  25. [33]

    Ptq4vit: Post-training quantization for vision transformers with twin uniform quantization

    Zhihang Yuan, Chenhao Xue, Yiqi Chen, Qiang Wu, and Guangyu Sun. Ptq4vit: Post-training quantization for vision transformers with twin uniform quantization. InECCV, pages 191–207, 2022. 2, 3, 6, 7

    2022

  26. [34]

    Cat-det: Con- trastively augmented transformer for multi-modal 3d object detection

    Yanan Zhang, Jiaxin Chen, and Di Huang. Cat-det: Con- trastively augmented transformer for multi-modal 3d object detection. InCVPR, 2022. 1

    2022

  27. [35]

    Transforming vision transformer: Towards efficient multi-task asynchronous learner

    Hanwen Zhong, Jiaxin Chen, Yutong Zhang, Di Huang, and Yunhong Wang. Transforming vision transformer: Towards efficient multi-task asynchronous learner. InNeurIPS, 2024. 1

    2024

  28. [36]

    I&s-vit: An inclusive & stable method for pushing the limit of post-training vits quantization.arXiv preprint arXiv:2311.10126, 2023

    Yunshan Zhong, Jiawei Hu, Mingbao Lin, Mengzhao Chen, and Rongrong Ji. I&s-vit: An inclusive & stable method for pushing the limit of post-training vits quantization.arXiv preprint arXiv:2311.10126, 2023. 6, 7

    2023

  29. [37]

    ERQ: Error reduction for post-training quanti- zation of vision transformers

    Yunshan Zhong, Jiawei Hu, You Huang, Yuxin Zhang, and Rongrong Ji. ERQ: Error reduction for post-training quanti- zation of vision transformers. InICML, 2024. 6, 7

    2024

  30. [38]

    Octr: Octree-based transformer for 3d object detection

    Chao Zhou, Yanan Zhang, Jiaxin Chen, and Di Huang. Octr: Octree-based transformer for 3d object detection. InCVPR,

  31. [39]

    Deformable DETR: deformable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable DETR: deformable transformers for end-to-end object detection. InICLR, 2021. 1 FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation Supplementary...

    2021

Pith tools

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