Pith. sign in

REVIEW 3 major objections 5 minor 63 references

Mix-QSAM: Mixed-Precision Quantization of the Segment Anything Model

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A static mixed-precision quantizer can match learning-based PTQ on SAM by allocating bits via KL importance and cross-layer synergy.

desk verdict Useful engineering results for SAM quantization, but the central synergy metric is mislabeled and the abstract overstates consistency. read the letter →

arxiv 2505.04861 v1 pith:3DGODW4H submitted 2025-05-08 cs.CV

classification cs.CV
keywords mixed-precisionquantizationpost-trainingSegmentAnythingModellayerimportanceKLdivergencecausalmutualinformationcross-layersynergyintegerquadraticprogramming
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

Mix-QSAM tests a straightforward hypothesis: for a prompt-based segmentation model like SAM, low-bit accuracy can be recovered without retraining or reconstruction—you only need to spend bits where they matter. The paper builds a static post-training quantizer that assigns each layer its own weight and activation bit-width by solving a small integer quadratic program. The optimization is driven by two scores: a per-layer importance score derived by zeroing that layer's activations and measuring the Kullback-Leibler divergence of the output, and a cross-layer synergy score that keeps adjacent interdependent layers at similar precision. On COCO instance segmentation and object detection, the method reports 17–20% higher average precision than fixed-bit post-training quantization at 4- and 6-bit settings, matching or beating learning-based reconstruction methods without the reconstruction cost.

What carries the argument

The load-bearing object is the pair of scores that feed the optimizer. Layer importance uses causal mutual information: zero the activations of layer $l$, measure $D_{\mathrm{KL}}(P_Y(y\mid L_l,L_{\neg l})\,\|\,P_Y(y\mid L_{\neg l}))$, normalize by the total over layers, and average over 64 calibration images to get $\Omega_l$. Cross-layer synergy approximates interdependence by $\hat{S}_{l,m}=\log\bigl(1+\bar{S}_{l,m}\bigr)$, where $\bar{S}_{l,m}$ is the mean over images of $1/(|I(L_l;Y)-I(L_m;Y)|+\epsilon)$. These two metrics enter a one-hot integer quadratic program: maximize $\sum_l \Omega_l\sum_j \alpha_{l,j} b_j - \lambda \sum_l \hat{S}_{l,m}\,|\sum_j \alpha_{l,j} b_j - \sum_j \alpha_{m,j} b_j|$ subject to exactly one bit-width per layer, a model-size budget, and a bit-operation budget. The optimization is what converts causal-ish scores into a concrete bit assignment for every quantized layer.

What would settle it

Quantize each encoder layer one at a time to 4-bit while keeping every other layer at full precision, and correlate the resulting mAP drop with that layer's importance score $\Omega_l$; if low-importance layers lose as much mAP as high-importance layers, or if reshuffling the bit assignment within the same total model size and bit-operation budget changes mAP by less than a point, then the importance and synergy metrics are not carrying the reported gains.

Watch

Extended reading notes

Core claim

The paper's central claim is that for SAM, per-layer bit-widths can be chosen before quantization, without retraining or reconstruction, by solving a small constrained integer quadratic program. The program's objective is a weighted sum of two causal-information metrics: a layer-importance score $\Omega_l$, computed by zeroing that layer's activations and measuring the KL divergence between the perturbed and unperturbed output distribution, averaged over 64 images; and a cross-layer synergy score $\hat{S}_{l,m}$, the log-scaled reciprocal of the difference between adjacent layers' importance scores. Maximizing $\sum_l \Omega_l\, b_l$ while penalizing $\hat{S}_{l,m}\,|b_l-b_m|$ assigns high precision to critical layers and keeps interdependent layers at similar precision. On COCO instance segmentation and object detection, the paper reports that this static allocation raises average precision by up to 20% in the abstract (17% in the contributions list) over fixed-bit PTQ at 6-bit and 4-bit settings, and matches or beats reconstruction-based methods while requiring no reconstruction. The claim is specifically about placement of bits, not about introducing new quantization operators.

Load-bearing premise

The method assumes that zeroing a layer's activations and measuring the KL divergence of the output distribution faithfully estimates that layer's causal contribution; because the output distribution in Eq. (14) is never explicitly defined, the importance scores that drive the entire bit allocation rest on an unspecified probability space.

Editorial extensions

If this is right

  • At 6-bit on the larger SAM variants, the mixed-precision static model ties full-precision mAP on several detector/model combinations, so a quantized SAM with no visible segmentation loss is within reach.
  • At 4-bit, the static method improves on the best reconstruction baseline by a wide margin on at least one SAM-B detector pairing (30.6 mAP versus 14.4 mAP), showing that reconstruction is not required to recover low-bit SAM.
  • The synergy penalty is worth 5.5–6.0 points of mAP in the SAM-B ablation, so abrupt bit changes between adjacent layers are a measurable source of quantization error.
  • Calibration stays practical: about 15 minutes for SAM-B, and no 200–400 minute reconstruction phase is needed.

Reading between the lines

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

  • Because the two scores are computed once on 64 images and the bit assignment is solved offline, the same allocation would likely transfer to other prompt distributions or downstream tasks without recalibration; the paper only demonstrates this on COCO instance segmentation and object detection.
  • The synergy penalty smooths a one-dimensional chain of bit widths. A natural test the paper does not run is to apply the same causal-information construction at finer granularity, such as attention heads or individual blocks, or to allow non-adjacent dependencies.
  • The reported 4-bit results suggest that bit placement, not rounding or reconstruction, is the dominant source of quantization error for SAM. If this interpretation is right, future PTQ for prompt-based models should focus on allocation metrics rather than reconstruction iterations, but the paper does not isolate that claim.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript proposes Mix-QSAM, a post-training mixed-precision quantization framework for the Segment Anything Model. It introduces a layer-wise importance score computed as a KL divergence between the original output distribution and the output distribution after zeroing a layer's activations, and a cross-layer synergy score intended to capture interaction information between adjacent layers. These scores are used in an integer quadratic programming formulation that allocates bit-widths under model-size and bit-operation constraints. Experiments on COCO instance segmentation and object detection with four detectors compare Mix-QSAM against static and learning-based PTQ baselines at W6/A6 and W4/A4 settings, and an ablation study isolates the effects of the importance score and the bit-transition regularizer.

Significance. If the proposed metrics are valid and the reported results are reproducible, this is a practically significant result: a static, reconstruction-free mixed-precision PTQ method that can match or exceed learning-based PTQ for SAM at a fraction of the calibration cost. The evaluation is not circular in an obvious way, because the importance and synergy scores are computed from 64 calibration images while mAP is measured on held-out COCO data, and the comparison set is reasonably broad. However, the central formal gaps in the definitions of both metrics and a constraint-feasibility inconsistency in the W4/A4 experiments mean that the significance is conditional on repair.

major comments (3)
  1. [§3.3, Eqs. (17)–(20) and Table 3] Equation (18) is presented as an approximation to the interaction information in Eq. (17), but no derivation is provided, and the printed denominator contains the typo |I(Lm;Yt)-I(Lm;Yt)|, which makes S_t_l_m equal to 1/epsilon for every pair. Under that literal reading, the BTR term in Eq. (21) is a constant-weight smoothness penalty, so the improvement of "Synergistic BTR" over "Independent BTR" in Table 3 (e.g., 29.3 to 35.3 for SAM-B W6/A6) cannot be attributed to cross-layer synergy. Even after correcting the denominator to |I(Ll;Yt)-I(Lm;Yt)|, the inverse difference of marginal importance scores is not a special case of Eq. (17) without additional assumptions; please provide the derivation or explicitly reclassify S as a heuristic and validate it separately.
  2. [§3.2, Eqs. (12)–(14)] The importance score rests on a formal identification that is not established. Equation (12) is the conditional mutual information I(Y;Li|L¬i), while Eq. (13) states an approximation by a KL divergence. These quantities coincide only after averaging the KL over the joint distribution of Li and L¬i; the paper does not write this expectation, and when the perturbation is implemented by zeroing activations, P(Y|L¬i) is an interventional distribution rather than the conditional distribution of the original model. The sample space of Y is also never defined: SAM's output is a mask or per-pixel probability map, so the summation over y in Eq. (14) is non-operational as written. Please define Y, the perturbed model, and the averaging procedure, and state precisely which quantity Omega_i is intended to estimate.
  3. [§3.4, Eqs. (21)–(24); §4; Figure 1] The statement in §4 that the mixed-precision models "match their fixed-precision counterparts in model size and bit-operations" is inconsistent with the candidate set shown in Figure 1. For W4/A4, if B = {4,5,6,7,8}, the only assignment satisfying a model-size constraint equal to the all-4-bit model size is the uniform all-4 assignment, because no bit-width below 4 is available to compensate for layers assigned above 4. In that case Mix-QSAM should coincide with the fixed-bit PTQ4SAM-S baseline, contradicting the large W4/A4 gains in Table 3 (e.g., 17.4 to 26.9 for SAM-B with YOLOX). Please report the actual candidate sets and constraint values used for each W/A setting, and clarify whether weights and activations are quantized with separate bit-widths, as the text in §3.4 suggests.
minor comments (5)
  1. [§4.4, Table 3] The ablation rows are labeled "Mix-QViT(ours)*" instead of "Mix-QSAM(ours)*"; this appears to be leftover text from a related method and should be corrected.
  2. [Abstract and §1] The abstract states "up to 20% higher average precision" while the Introduction states "up to 17% higher"; please specify whether these are absolute mAP point differences or relative improvements and harmonize the numbers.
  3. [§3.3, text after Eq. (18)] The sentence defining the denominator writes |I(Lm;Yt)-I(Lm;Yt)|; this should be |I(Ll;Yt)-I(Lm;Yt)|.
  4. [Figure 1 caption] The caption reads "SAM- under W6/A6"; the model name is incomplete, and it should also state whether the candidate set {4,5,6,7,8} is the one used for W4/A4 experiments.
  5. [§3.4, Eq. (21)] Equation (21) uses a single bit-width variable b_j for each layer, while the text in §3.4 says the method assigns different bit-widths (bw_l, ba_l) to weights and activations; please clarify the relationship between the two notations.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the importance and synergy scores are precomputed heuristics and the reported gains are empirical; the Eq. (18) approximation is a validity defect, not a circular reduction.

full rationale

The derivation chain is not circular in the forbidden sense. The layer importance score (Section 3.2, Eqs. 12-16) is computed by zeroing activations on calibration images and measuring a KL shift; it is not fitted to the COCO mAP values reported later. The mixed-precision IQP (Section 3.4, Eq. 21) uses these precomputed scores as inputs and is evaluated on held-out data, so there is no fitted-input-called-prediction pattern. PTQ4SAM, used as the quantization backbone, is an external baseline, and the self-citations to LRP-QViT and Mix-QViT are contextual, not load-bearing: the paper does not justify its central metric by citing those works. The weak point is Section 3.3: Eq. (18) does not follow from Eq. (17), and the typo '|I(Lm;Yt)-I(Lm;Yt)|' would make the synergy score constant; even with the intended correction, the inverse-difference heuristic is an unjustified approximation rather than a derivation. Similarly, PY(y) in Eq. (14) is never operationalized. These are correctness and soundness defects, not cases where a prediction equals its input by construction, so they do not raise the circularity score beyond the minor self-citation/naming inconsistency in Table 3, where the method is mislabeled 'Mix-QViT(ours)*'.

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

The central claim rests on inferred quantities that are never measured or validated against a ground-truth notion of layer importance or inter-layer dependence. The two metrics, a perturbation-based KL score and an inverse-difference synergy, are the entire basis for the IQP allocation; neither is derived from the information-theoretic quantities named in the text, and neither has an independent falsifiable handle.

free parameters (5)
  • lambda (bit transition penalty weight) = 0.1
    Eq. (21), chosen by hand with no sensitivity analysis; it controls the entire synergy regularization effect.
  • epsilon (synergy denominator offset) = unspecified
    Eq. (18), small constant preventing division by zero; value never given.
  • candidate bit-width set B = {4, 5, 6, 7, 8}
    Figure 1 caption; the allowed bit-widths are chosen ad hoc and not justified.
  • calibration and importance sample counts = 32 calibration images, T=64 for importance averaging
    Sections 3.2 and 4; the counts are arbitrary and no sensitivity or variance is reported.
  • model size and bit-operation budgets (CM, CBOP) = not specified numerically
    Section 3.4; the budgets are said to match fixed-bit baselines, so they vary per experiment and are not reported.
assumptions (5)
  • domain assumption The KL divergence between the output distributions of the original and activation-zeroed models can be estimated and averaged over 64 images.
    Section 3.2, Eqs. (13)-(16): the paper never defines the sample space of PY(y) for SAM's dense per-pixel outputs, so the entropy and KL terms are non-operational.
  • domain assumption Zeroing a layer's activations is a valid causal intervention that isolates the layer's contribution to the output.
    Section 3.2: the method treats activation zeroing as causal elimination, but in a residual network this removes only the layer's residual delta, not the entire contribution path.
  • ad hoc to paper Cross-layer synergy in Eq. (18) is an acceptable approximation of the joint mutual information synergy in Eq. (17).
    No derivation connects the inverse difference of individual importance scores to I(Ll,Lm;Y) - I(Ll;Y) - I(Lm;Y); the approximation is asserted and, as printed, degenerates.
  • domain assumption The KL-based layer importance and the synergy penalty in Eq. (21) are accurate surrogates for quantization-induced mAP loss.
    The objective maximizes a weighted bit-width sum, not a direct accuracy estimate; the link between the importance score and final task accuracy is assumed.
  • standard math Integer quadratic programming via CVXPY finds a useful optimum of Eq. (21).
    The problem is a binary quadratic program with linear constraints; standard solvers can handle it, but no optimality certificate or solver-specific configuration is provided.
invented entities (2)
  • Cross-layer synergy metric (S_hat)
    purpose: Regularizes the IQP to assign similar bit-widths to adjacent layers based on their claimed interdependence.
    Presented as a causal mutual information synergy, but implemented as an inverse difference of individual importance scores; no external validation links it to actual inter-layer dependence.
  • Causal mutual information importance score (Omega)
    purpose: Ranks each layer's contribution to SAM's output for bit-width allocation.
    Named as causal mutual information, but computed as a perturbation-based KL divergence between output maps; the causal interpretation is unsupported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mix-QSAM: Mixed-Precision Quantization of the Segment Anything Model." pith.science (2026). https://pith.science/paper/3DGODW4H

@misc{pith2026250504861,
  author       = {Pith},
  title        = {Pith review of: Mix-QSAM: Mixed-Precision Quantization of the Segment Anything Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3DGODW4H}},
  note         = {Machine review of arXiv:2505.04861}
}
read the original abstract

The Segment Anything Model (SAM) is a popular vision foundation model; however, its high computational and memory demands make deployment on resource-constrained devices challenging. While Post-Training Quantization (PTQ) is a practical approach for reducing computational overhead, existing PTQ methods rely on fixed bit-width quantization, leading to suboptimal accuracy and efficiency. To address this limitation, we propose Mix-QSAM, a mixed-precision PTQ framework for SAM. First, we introduce a layer-wise importance score, derived using Kullback-Leibler (KL) divergence, to quantify each layer's contribution to the model's output. Second, we introduce cross-layer synergy, a novel metric based on causal mutual information, to capture dependencies between adjacent layers. This ensures that highly interdependent layers maintain similar bit-widths, preventing abrupt precision mismatches that degrade feature propagation and numerical stability. Using these metrics, we formulate an Integer Quadratic Programming (IQP) problem to determine optimal bit-width allocation under model size and bit-operation constraints, assigning higher precision to critical layers while minimizing bit-width in less influential layers. Experimental results demonstrate that Mix-QSAM consistently outperforms existing PTQ methods on instance segmentation and object detection tasks, achieving up to 20% higher average precision under 6-bit and 4-bit mixed-precision settings, while maintaining computational efficiency.

Figures

Figures reproduced from arXiv: 2505.04861 by the authors.

Figure 1
Figure 1. Mixed-precision bit allocation for SAM- under W6/A6: comparing layer importance score alone vs. using both layer importance [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Qualitative comparison of instance segmentation on the COCO dataset using a 4-bit Segment Anything Model (SAM) with [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 45 canonical work pages

  1. [1]

    Crowd-SAM: Sam as a smart annotator for object detection in crowded scenes

    Zhi Cai, Yingjie Gao, Yaoyan Zheng, Nan Zhou, and Di Huang. Crowd-SAM: Sam as a smart annotator for object detection in crowded scenes. In Eur. Conf. Comput. Vis. (ECCV), pages 334–351, 2024. 1

  2. [2]

    DearKD: Data-efficient early knowledge distillation for vision transformers

    Xianing Chen et al. DearKD: Data-efficient early knowledge distillation for vision transformers. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 12052–12062, 2022. 1

  3. [3]

    An effective in- formation theoretic framework for channel pruning

    Yihao Chen and Zefang Wang. An effective in- formation theoretic framework for channel pruning. arXiv:2408.16772, 2024. 3

  4. [4]

    Tracking any- thing with decoupled video segmentation

    Ho Kei Cheng, Seoung Wug Oh, Brian Price, Alexan- der Schwing, and Joon-Young Lee. Tracking any- thing with decoupled video segmentation. In Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , pages 1316–1326, 2023. 1, 2

  5. [5]

    Low-bit quantization of neural networks for efficient inference

    Yoni Choukroun, Eli Kravchik, Fan Yang, and Pavel Kisilev. Low-bit quantization of neural networks for efficient inference. In Proc. IEEE/CVF Int. Conf. Comput. Vis. Worksh. (ICCVW) , pages 3009–3018. IEEE, 2019. 6, 7

  6. [6]

    Mixed-Precision Quantization for Deep Vision Models with Integer Quadratic Programming

    Zihao Deng, Xin Wang, Sayeh Sharify, and Michael 9 Orshansky. Mixed-precision quantization with cross- layer dependencies. arXiv:2307.05657, 2023. 2, 3, 5

  7. [7]

    CVXPY: A python-embedded modeling language for convex op- timization

    Steven Diamond and Stephen Boyd. CVXPY: A python-embedded modeling language for convex op- timization. Journal of Machine Learning Research , 17(83):1–5, 2016. 6

  8. [8]

    Towards accurate post-training quan- tization for vision transformer

    Yifu Ding, Haotong Qin, Qinghua Yan, Zhenhua Chai, Junjie Liu, et al. Towards accurate post-training quan- tization for vision transformer. In Proc. 30th ACM Int. Conf. Multimedia, pages 5380–5388, 2022. 1

Show all 63 references
  1. [9]

    HAWQ: Hessian aware quantization of neural networks with mixed-precision

    Zhen Dong, Zhewei Yao, Amir Gholami, Michael W Mahoney, and Kurt Keutzer. HAWQ: Hessian aware quantization of neural networks with mixed-precision. In Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV)) , pages 293–302, 2019. 2, 3, 4, 5

  2. [10]

    HAWQ- V2: Hessian aware trace-weighted quantization of neural networks

    Zhen Dong, Zhewei Yao, Daiyaan Arfeen, Amir Gho- lami, Michael W Mahoney, and Kurt Keutzer. HAWQ- V2: Hessian aware trace-weighted quantization of neural networks. Adv. Neural Inform. Process. Syst. (NeurIPS), 33:18518–18529, 2020. 2, 3, 4, 5

  3. [11]

    Layer-wise model pruning based on mutual information

    Chun Fan, Jiwei Li, Xiang Ao, Fei Wu, Yuxian Meng, and Xiaofei Sun. Layer-wise model pruning based on mutual information. arXiv:2108.12594, 2021. 3

  4. [12]

    YOLOX: Exceeding yolo series in 2021

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. YOLOX: Exceeding yolo series in 2021. arXiv:2107.08430, 2021. 6, 7, 8

  5. [13]

    Soft filter pruning for accelerating deep convolutional neural networks

    Yang He, Guoliang Kang, Xuanyi Dong, Yanwei Fu, and Yi Yang. Soft filter pruning for accelerating deep convolutional neural networks. In Proc. 27th Int. Joint Conf. Artif. Intell. (IJCAI), pages 2234–2240, 2018. 1

  6. [14]

    Quantization and train- ing of neural networks for efficient integer-arithmetic- only inference

    Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, et al. Quantization and train- ing of neural networks for efficient integer-arithmetic- only inference. In Proc. IEEE Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 2704–2713, 2018. 6, 7

  7. [15]

    Detrs with hybrid matching

    Ding Jia, Yuhui Yuan, Haodi He, Xiaopei Wu, Hao- jun Yu, et al. Detrs with hybrid matching. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 19702–19712, 2023. 6, 7

  8. [16]

    Segment anything in high quality

    Lei Ke, Mingqiao Ye, Martin Danelljan, Yu-Wing Tai, Chi-Keung Tang, Fisher Yu, et al. Segment anything in high quality. Adv. Neural Inform. Process. Syst. (NeurIPS), 36:29914–29934, 2023. 1, 2

  9. [17]

    SAM-Net: self-attention based feature matching with spatial transformers and knowledge distillation

    Benjamin Kelenyi, Victor Domsa, and Levente Tamas. SAM-Net: self-attention based feature matching with spatial transformers and knowledge distillation. Ex- pert Systems with Applications, 242:122804, 2024. 1, 2

  10. [18]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, et al. Segment anything. InProc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , pages 4015–4026, 2023. 1, 6

  11. [19]

    Optimizing expo- nent bias for sub-8bit floating-point inference of fine- tuned transformers

    Janghwan Lee and Jungwook Choi. Optimizing expo- nent bias for sub-8bit floating-point inference of fine- tuned transformers. In IEEE 4th Int. Conf. Artif. Intell. Circuits and Syst. (AICAS), pages 98–101, 2022. 5

  12. [20]

    FlexRound: Learnable rounding based on element-wise division for post-training quantiza- tion

    Jung Hyun Lee, Jeonghoon Kim, Se Jung Kwon, and Dongsoo Lee. FlexRound: Learnable rounding based on element-wise division for post-training quantiza- tion. In Int. Conf. Machine Learn. (ICML) , pages 18913–18939, 2023. 3

  13. [21]

    Differ- entiable search for finding optimal quantization strat- egy

    Lianqiang Li, Chenqian Yan, and Yefei Chen. Differ- entiable search for finding optimal quantization strat- egy. arXiv:2404.08010, 2024. 3

  14. [22]

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

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, et al. BRECQ: Pushing the limit of post-training quantization by block reconstruction. arXiv:2102.05426, 2021. 3, 6, 7, 8

  15. [23]

    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. In Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , pages 17227–17236, 2023. 1, 3, 4

  16. [24]

    Knowledge distillation via the target- aware transformer

    Sihao Lin et al. Knowledge distillation via the target- aware transformer. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 10915–10924,

  17. [25]

    Microsoft COCO: Com- mon objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, et al. Microsoft COCO: Com- mon objects in context. In Computer vision–ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceedings, part v 13 , pages 740–755, 2014. 6, 7, 8

  18. [26]

    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. In Proc. 31st Int. Joint Conf. Artif. Intell. (IJCAI) , pages 1173– 1179, 2022. 1, 3

  19. [27]

    Perceptual-sensitive gan for generating adversarial patches

    Aishan Liu, Xianglong Liu, Jiaxin Fan, Yuqing Ma, Anlan Zhang, Huiyuan Xie, and Dacheng Tao. Perceptual-sensitive gan for generating adversarial patches. In Proc. AAAI Conf. Artif. Intell. (AAAI) , pages 1028–1035, 2019. 1, 2

  20. [28]

    PD-Quant: Post- training quantization based on prediction difference metric

    Jiawei Liu, Lin Niu, Zhihang Yuan, Dawei Yang, Xinggang Wang, and Wenyu Liu. PD-Quant: Post- training quantization based on prediction difference metric. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 24427–24437, 2023. 3, 6

  21. [29]

    PQ-SAM: Post-training quantization for segment anything model

    Xiaoyu Liu, Xin Ding, Lei Yu, Yuanyuan Xi, Wei Li, Zhijun Tu, Jie Hu, Hanting Chen, Baoqun Yin, and Zhiwei Xiong. PQ-SAM: Post-training quantization for segment anything model. In Eur. Conf. Comput. Vis. (ECCV), pages 420–437. Springer, 2024. 1, 2 10

  22. [30]

    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. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR) , pages 20321–20330, 2023. 1

  23. [31]

    Post-training quantization for vision transformer

    Zhenhua Liu, Yunhe Wang, Kai Han, Wei Zhang, Si- wei Ma, and Wen Gao. Post-training quantization for vision transformer. Adv. Neural Inform. Process. Syst. (NeurIPS), 34:28092–28103, 2021. 2

  24. [32]

    AutoQ: Automated kernel-wise neural network quantization

    Qian Lou, Feng Guo, Minje Kim, Lantao Liu, and Lei Jiang. AutoQ: Automated kernel-wise neural network quantization. In Int. Conf. Learn. Represent. (ICLR) ,

  25. [33]

    DeepBurning-MixQ: An open source mixed-precision neural network accelerator de- sign framework for fpgas

    Erjing Luo, Haitong Huang, Cheng Liu, Guoyu Li, Bing Yang, et al. DeepBurning-MixQ: An open source mixed-precision neural network accelerator de- sign framework for fpgas. In 2023 IEEE/ACM Inter- national Conference on Computer Aided Design (IC- CAD), pages 1–9, 2023. 3

  26. [34]

    PTQ4SAM: Post-training quan- tization for segment anything

    Chengtao Lv, Hong Chen, Jinyang Guo, Yifu Ding, and Xianglong Liu. PTQ4SAM: Post-training quan- tization for segment anything. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 15941–15951, 2024. 1, 2, 3, 4, 6, 7, 8

  27. [35]

    Seg- ment anything model for medical image analysis: an experimental study

    Maciej A Mazurowski, Haoyu Dong, Hanxue Gu, Jichen Yang, Nicholas Konz, and Yixin Zhang. Seg- ment anything model for medical image analysis: an experimental study. Medical Image Analysis , 89: 102918, 2023. 1

  28. [36]

    SAM-PM: Enhancing video camouflaged ob- ject detection using spatio-temporal attention

    Muhammad Nawfal Meeran, Bhanu Pratyush Mantha, et al. SAM-PM: Enhancing video camouflaged ob- ject detection using spatio-temporal attention. InProc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 1857–1866, 2024. 1

  29. [37]

    Up or down? adaptive rounding for post-training quantiza- tion

    Markus Nagel, Rana Ali Amjad, Mart Van Baalen, Christos Louizos, and Tijmen Blankevoort. Up or down? adaptive rounding for post-training quantiza- tion. In Int. Conf. Machine Learn. (ICML) , pages 7197–7206. PMLR, 2020. 3, 6, 7

  30. [38]

    LRP-QViT: Mixed-precision vision transformer quantization via layer-wise relevance propagation

    Navin Ranjan and Andreas Savakis. LRP-QViT: Mixed-precision vision transformer quantization via layer-wise relevance propagation. arXiv:2401.11243,

  31. [39]

    Mix-QViT: Mixed-precision vision transformer quantization driven by layer importance and quantization sensitiv- ity

    Navin Ranjan and Andreas Savakis. Mix-QViT: Mixed-precision vision transformer quantization driven by layer importance and quantization sensitiv- ity. arXiv:2501.06357, 2025. 1, 2, 3, 4, 5

  32. [40]

    Faster R-CNN: Towards real-time object detec- tion with region proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster R-CNN: Towards real-time object detec- tion with region proposal networks. Adv. Neural In- form. Process. Syst. (NeurIPS), 28, 2015. 6, 7

  33. [41]

    Quantized-ViT efficient training via fisher matrix regularization

    Yuzhang Shang, Gaowen Liu, Ramana Kompella, and Yan Yan. Quantized-ViT efficient training via fisher matrix regularization. In International Conference on Multimedia Modeling , pages 270–284. Springer,

  34. [42]

    Anything-3D: Towards single-view anything recon- struction in the wild

    Qiuhong Shen, Xingyi Yang, and Xinchao Wang. Anything-3D: Towards single-view anything recon- struction in the wild. arXiv:2304.10261, 2023. 1, 2

  35. [43]

    TinySAM: Pushing the envelope for efficient segment anything model

    Han Shu, Wenshuo Li, Yehui Tang, Yiman Zhang, Yi- hao Chen, et al. TinySAM: Pushing the envelope for efficient segment anything model. arXiv:2312.13789,

  36. [44]

    MPTQ-ViT: Mixed-precision post-training quantization for vision transformer

    Yu-Shan Tai et al. MPTQ-ViT: Mixed-precision post-training quantization for vision transformer. arXiv:2401.14895, 2024. 4, 5

  37. [45]

    Mixed-precision neural network quantization via learned layer-wise importance

    Chen Tang, Kai Ouyang, Zhi Wang, Yifei Zhu, Wen Ji, et al. Mixed-precision neural network quantization via learned layer-wise importance. In Eur. Conf. Comput. Vis. (ECCV), pages 259–275, 2022. 5

  38. [46]

    HAQ: Hardware-aware automated quantization with mixed precision

    Kuan Wang, Zhijian Liu, Yujun Lin, Ji Lin, and Song Han. HAQ: Hardware-aware automated quantization with mixed precision. In Proc. IEEE/CVF Conf. Com- put. Vis. and Pattern Recog. (CVPR) , pages 8612– 8620, 2019. 2

  39. [47]

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

    Xiuying Wei, Ruihao Gong, Yuhang Li, Xianglong Liu, and Fengwei Yu. Qdrop: Randomly dropping quantization for extremely low-bit post-training quan- tization. arXiv:2203.05740, 2022. 3, 6, 7, 8

  40. [48]

    Mutual information preserving neural network pruning

    Charles Westphal, Stephen Hailes, and Mirco Mu- solesi. Mutual information preserving neural network pruning. arXiv:2411.00147, 2024. 3

  41. [49]

    Integer quantization for deep learning inference: Principles and empirical evaluation

    Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev, and Paulius Micikevicius. Integer quantization for deep learning inference: Principles and empirical evaluation. arXiv:2004.09602, 2020. 6, 7

  42. [50]

    Medical sam adapter: Adapting segment anything model for medical image segmentation

    Junde Wu, Wei Ji, Yuanpei Liu, Huazhu Fu, Min Xu, Yanwu Xu, and Yueming Jin. Medical sam adapter: Adapting segment anything model for medical image segmentation. arXiv:2304.12620, 2023. 1, 2

  43. [51]

    AdaLog: Post-training quantization for vision transformers with adaptive log- arithm quantizer

    Zhuguanyu Wu, Jiaxin Chen, Hanwen Zhong, Di Huang, and Yunhong Wang. AdaLog: Post-training quantization for vision transformers with adaptive log- arithm quantizer. In Eur. Conf. Comput. Vis. (ECCV), pages 411–427, 2025. 1, 3, 4

  44. [52]

    Patch-wise mixed-precision quantization of vi- sion transformer

    Junrui Xiao, Zhikai Li, Lianwei Yang, and Qingyi Gu. Patch-wise mixed-precision quantization of vi- sion transformer. 2023 Int. Joint Conf. Neural Net- works (IJCNN), pages 1–7, 2023. 2, 4

  45. [53]

    EfficientSAM: Lever- aged masked image pretraining for efficient segment anything

    Yunyang Xiong, Bala Varadarajan, Lemeng Wu, Xi- aoyu Xiang, Fanyi Xiao, et al. EfficientSAM: Lever- aged masked image pretraining for efficient segment anything. In Proc. IEEE/CVF Conf. Comput. Vis. and 11 Pattern Recog. (CVPR), pages 16111–16121, 2024. 1, 2

  46. [54]

    Mixed precision quantization of transformer language models for speech recognition

    Junhao Xu, Shoukang Hu, Jianwei Yu, Xunying Liu, and Helen Meng. Mixed precision quantization of transformer language models for speech recognition. In IEEE Int. Conf. Acoust., Speech and Sig Process. (ICASSP), pages 7383–7387, 2021. 5

  47. [55]

    Global vi- sion transformer pruning with hessian-aware saliency

    Huanrui Yang, Hongxu Yin, Maying Shen, Pavlo Molchanov, Hai Li, and Jan Kautz. Global vi- sion transformer pruning with hessian-aware saliency. In Proc. IEEE/CVF Conf. Comput. Vis. and Pattern Recog. (CVPR), pages 18547–18557, 2023. 2, 3

  48. [56]

    Track Anything: Segment anything meets videos

    Jinyu Yang, Mingqi Gao, Zhe Li, Shang Gao, Fangjing Wang, and Feng Zheng. Track Anything: Segment anything meets videos. arXiv:2304.11968,

  49. [57]

    Width & depth pruning for vision transformers

    Fang Yu, Kun Huang, Meng Wang, Yuan Cheng, Wei Chu, and Li Cui. Width & depth pruning for vision transformers. In Proc. AAAI Conf. Artif. Intell. (AAAI), pages 3143–3151, 2022. 1

  50. [58]

    Inpaint Anything: Segment anything meets image inpainting

    Tao Yu, Runseng Feng, Ruoyu Feng, Jinming Liu, Xin Jin, et al. Inpaint Anything: Segment anything meets image inpainting. arXiv:2304.06790, 2023. 1, 2

  51. [59]

    PTQ4ViT: Post-training quantiza- tion for vision transformers with twin uniform quan- tization

    Zhihang Yuan, Chenhao Xue, Yiqi Chen, Qiang Wu, and Guangyu Sun. PTQ4ViT: Post-training quantiza- tion for vision transformers with twin uniform quan- tization. In Eur. Conf. Comput. Vis. (ECCV) , pages 191–207, 2022. 1, 3, 6

  52. [60]

    Faster Segment Any- thing: Towards lightweight sam for mobile applica- tions

    Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, et al. Faster Segment Any- thing: Towards lightweight sam for mobile applica- tions. arXiv:2306.14289, 2023. 1, 2

  53. [61]

    DINO: DETR with improved denois- ing anchor boxes for end-to-end object detection

    Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, et al. DINO: DETR with improved denois- ing anchor boxes for end-to-end object detection. arXiv:2203.03605, 2022. 6, 7

  54. [62]

    Fast segment anything

    Xu Zhao, Wenchao Ding, Yongqi An, Yinglong Du, Tao Yu, Min Li, Ming Tang, and Jinqiao Wang. Fast segment anything. arXiv:2306.12156, 2023. 1, 2

  55. [63]

    DarkSAM: Fooling segment anything model to segment nothing

    Ziqi Zhou, Yufei Song, Minghui Li, Shengshan Hu, Xianlong Wang, et al. DarkSAM: Fooling segment anything model to segment nothing. Adv. Neural Inform. Process. Syst. (NeurIPS) , 37:49859–49880,

Pith tools

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