Pith. sign in

REVIEW 4 major objections 5 minor 63 references

Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that channel pruning for semantic segmentation should be driven by spatial redundancy in feature maps, not weight similarity, and that its SIRFP method achieves higher mIoU at matched FLOPs than prior pruning methods.

desk verdict Real novelty in the spatial-aware pruning metric, but the headline comparison is undermined by two inconsistent unpruned baselines that need to be reconciled. read the letter →

arxiv 2412.12672 v1 pith:LHIG4B7G submitted 2024-12-17 cs.CV

Dongyue Wu , Zilin Guo , Li Yu , Nong Sang , Changxin Gao This is my paper · ORCID
classification cs.CV MSC 68T0768T4568R1005C6905C85
keywords filterpruningsemanticsegmentationspatial-awareredundancymaximumedgeweightcliqueproblemgreedychannelfeature-mapmodelcompressionEMAstatistics
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

Most filter-pruning methods are built for image classification and score channels by weight similarity or pooled features, which discards where in the image a channel fires. This paper argues that semantic segmentation needs a spatial-aware redundancy measure, and proposes SIRFP, which compares full-resolution feature maps with the Jensen-Shannon divergence and prunes the channels whose spatial patterns are most duplicated. The paper claims that selecting kept channels to minimize pairwise redundancy is exactly a maximum-edge-weight clique problem, and that a greedy one-by-one deletion solver makes this practical even for layers with more than 1000 channels. If correct, the method means location-sensitive tasks need their own pruning metrics, and the reported results show higher mIoU at 60–80% FLOPs reduction than FPGM and DCFP on Cityscapes, with similar gains on ADE20K and COCO-Stuff.

What carries the argument

The engine is a complete edge-weighted graph $G=(V,E)$ whose vertices are the channels of a layer and whose edge weight $a_{ij}$ is the EMA-accumulated value of $1 - r_{ij}$, where $r_{ij}$ is the Jensen-Shannon divergence between the two channels' full-resolution feature maps (Equation 3). Pruning becomes the Maximum Edge Weight Clique Problem: choose the $b$ vertices whose induced subgraph has the largest total edge weight, which is equivalent to keeping the least mutually redundant channels. The efficient heuristic greedy pruning (EHGP) then repeatedly removes the vertex with the smallest current sum of incident edge weights, updating the remaining sums in $O(n^2)$ total time, and a global FLOPs threshold converts per-layer sums into a pruning decision across the network.

What would settle it

Re-run the Deeplabv3-ResNet50/Cityscapes pipeline with the edge-weight matrix accumulation started only after the first half of pre-training, or with the EMA hyperparameter alpha set to 0.9 instead of 0.99. If the 60%-FLOPs mIoU stays near 81.3 and the margin over FPGM and DCFP persists, the EMA-schedule assumption is not load-bearing; if the mIoU drops by more than about 0.5 points, the reported results depend on statistics gathered from an immature network.

Watch

Extended reading notes

Core claim

The central claim is that a channel is redundant not when its weights resemble another channel's weights, but when its output feature map, taken at full spatial resolution, carries information already present in other kept channels. SIRFP encodes this as edge weights $a_{ij}$ between every pair of channels, set to the EMA-accumulated complement of a Jensen-Shannon divergence between their un-pooled feature maps, and keeps the budget of channels whose total edge weight is maximal. The paper shows that this objective is the Maximum Edge Weight Clique Problem, that the one-vertex-deletion case has a closed-form solution, and that iterating that deletion gives a greedy solver with $O(n^2)$ average complexity. On Deeplabv3-ResNet50 with Cityscapes, the method reaches 81.3% mIoU at 60% FLOPs reduction, versus 80.2% for FPGM and 80.9% for DCFP, and the margin widens at higher compression ratios; the same pipeline also transfers to ADE20K, COCO-Stuff, object detection, and image classification.

Load-bearing premise

The load-bearing premise is that the running average of channel-redundancy scores, collected from the network while it is still being trained, accurately reflects which channels are redundant in the fully trained network; if early-training features are uninformative or the feature distribution shifts substantially, the pruning decisions would be based on stale statistics.

Editorial extensions

If this is right

  • At 60% FLOPs reduction on Deeplabv3-ResNet50 over Cityscapes, SIRFP yields 81.3% mIoU, 1.1 points above FPGM and 0.4 above DCFP with slightly fewer remaining FLOPs.
  • At 70% and 80% reductions the margins over DCFP grow to 1.1 and 0.6 mIoU, showing the spatial metric degrades more gracefully under aggressive compression.
  • The same EMA-accumulated edge weights transfer across tasks: 44.9% mIoU on ADE20K and 34.8% on COCO-Stuff at 60% reduction, 26.6 AP on SSD-ResNet50/COCO at 50% FLOPs cut, and 77.35% top-1 on ImageNet at 2.0G FLOPs.
  • Because redundancy statistics are accumulated during training via EMA, the pipeline avoids the separate feature-collection inference pass used by HRank- and CHIP-style pruning methods.

Reading between the lines

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

  • An untested extension is that the same spatial-redundancy objective should help other dense prediction tasks, such as monocular depth estimation or instance segmentation, where localization errors are directly penalized; running SIRFP on those tasks would separate the metric's contribution from the MEWCP solver's.
  • Their own runtime table shows the exact MEWCP solver reaches 81.4% mIoU versus 81.3% for the greedy EHGP, so a stronger local-search heuristic that looks two deletions ahead might capture most of the remaining 0.1-point gap without the solver's 10,800-second cost.
  • The paper fixes EMA alpha at 0.99 and starts accumulation at iteration 1 without ablation; a testable prediction is that beginning accumulation later, or using a smaller alpha, would shift which channels survive and either shrink or widen the gap to DCFP.
  • If feature-map redundancy patterns stabilize early in training, the method could be applied to an already-trained network with only a short feature-collection pass, which would remove the paper's own stated limitation that it must run during training.
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 SIRFP, a structural pruning method for semantic segmentation. The key ideas are: (i) measuring channel redundancy with a spatial-aware JS-divergence metric computed on full-resolution feature maps, accumulated via EMA during training; (ii) formulating the selection of retained channels as a Maximum Edge Weight Clique Problem (MEWCP); and (iii) solving it with a greedy heuristic (EHGP). Experiments on Cityscapes, ADE20K, COCO-stuff, COCO object detection, and ImageNet report favorable parameter/FLOPs versus accuracy trade-offs, including a claimed best parameter-mIoU trade-off on Cityscapes relative to pruning and real-time segmentation methods.

Significance. If the empirical claims hold, the spatial-aware redundancy formulation is a meaningful step for pruning location-sensitive models, and the greedy MEWCP solver is a practical contribution. The paper includes useful ablations (feature resolution, redundancy metric, solver runtime), reports results across multiple datasets and tasks, and releases code. The main concern is that the central Cityscapes comparison rests on inconsistent unpruned-baseline numbers, and several algorithmic details are stated ambiguously. These issues are fixable but currently prevent the results from being interpreted reliably.

major comments (4)
  1. [Table 1 vs Table 7 and Appendix C (Tables 8-9)] The unpruned Deeplabv3-ResNet50 baseline on Cityscapes is reported as 81.6 mIoU in Table 1 but as 79.3 mIoU in Table 7 and in Appendix C, Tables 8 and 9. The central claim that Ours-60% (81.3) outperforms FPGM (80.2) and DCFP (80.9) is made against the 81.6 baseline. If the correct baseline is 79.3, then Ours-60% is 2.0 points above the unpruned model, which cannot be explained by pruning alone; it implies that extra fine-tuning, GSRL, progressive pruning, or evaluation differences contribute to the gain. The manuscript must reconcile the two baseline numbers and add a no-pruning control that receives the same fine-tuning schedule as the pruned models before the advantage over FPGM and DCFP can be attributed to the spatial-aware metric. Table 7 also reports DCFP-60% as 80.7, whereas Table 1 reports 80.9, further indicating that the comparison pipeline is not stable across tables.
  2. [Algorithm 1 and Appendix D (global threshold)] Algorithm 1 is stated inconsistently with the surrounding text and with Appendix D. The text says the task is to delete C_l - b channels iteratively, but the loop in Algorithm 1 reads 'for each prune iteration t_p in [b]'; since b denotes the number of kept channels in Eq. (2), the loop bound should be C_l - b. In addition, Appendix D says that after collecting the sums s_k, the global threshold is set so that 'pruning all the channels in the whole network whose s_k >= s_th' satisfies the FLOPs target, whereas Algorithm 1 always removes the vertex with the minimum s_i. These two descriptions identify opposite sets of channels. The algorithm must be restated unambiguously for the method to be reproducible.
  3. [Algorithm 2 and Eq. (4)] Algorithm 2 accumulates the edge-weight matrix A^l via EMA starting from the first pre-training iteration, with alpha fixed at 0.99. Because early-training features are not representative of the final trained network, the running average may be dominated by stale statistics when pruning decisions are made. The paper does not ablate the EMA decay alpha or the accumulation start time, nor does it provide evidence that the running average converges to the final network's redundancy pattern. This design choice is load-bearing for the claimed benefit of the spatial-aware metric, and an ablation is needed to show that the reported results do not depend on a particular warm-up or decay value.
  4. [Table 5 and Appendix D (ImageNet schedules)] The ImageNet comparison in Table 5 is not controlled for training schedule. Appendix D reports that SIRFP is trained for 250 epochs with a cosine schedule, pruning every 2 epochs from epoch 2 to 180, while the paper does not report the schedules used for the listed baselines, which are typically trained for fewer epochs. A longer training schedule alone can improve accuracy, so the 77.35 top-1 accuracy at 2.0G in Table 5 cannot currently be attributed to the pruning criterion. The authors should either run all methods under the same schedule or provide the training schedule used for each baseline.
minor comments (5)
  1. [Eq. (3)] Equation (3) writes F_i^l, F_j^l as elements of R^{C_l x H x W}, but these are per-filter feature maps and should be R^{H x W} (or R^{1 x H x W}). The implementation in Appendix C states that features are normalized as probability score maps for Dice/KL/JS, but this normalization is not described in the main text and should be stated before Eq. (3).
  2. [Eq. (4)] The initial value of the EMA edge weight a^0_ij is not specified. If it is initialized differently (e.g., to 0 versus to the first observed value), the early updates and potentially the pruning decisions change; please specify the initialization.
  3. [Algorithm 2 / Limitation] The Limitation section says that for a well-trained network SIRFP must inference on training samples to acquire feature maps, but Algorithm 2 accumulates redundancy during pre-training. Please clarify when EMA accumulation is performed in the actual pipeline and whether the released code supports both settings.
  4. [Abstract / Introduction] The text repeatedly refers to 'mutual information redundancy', but the proposed metric is the Jensen-Shannon divergence between feature distributions (Eq. (3)). Please clarify the relationship between the two notions, since JS divergence is not mutual information.
  5. [Appendix D] The appendix contains a typo: 'Talbe 10' should be 'Table 10'. The phrase 'batch of size' should also be 'batch size'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the redundancy metric, MEWCP formulation, and greedy solver are defined independently of the reported mIoU results; the inconsistent unpruned baselines and validation-set ablations are correctness concerns, not circularity.

full rationale

The paper's derivation chain is self-contained rather than circular. The spatial-aware redundancy metric (Eq. 3) is defined on un-pooled feature maps using a standard JS-divergence-like expression, and the edge weights (Eq. 4) are accumulated by EMA during training; neither quantity is defined in terms of the mIoU numbers that the paper reports. The pruning objective is then formulated as a Maximum Edge Weight Clique Problem (Eq. 2), with the equivalence to minimizing pairwise redundancy holding by construction because edge weights are defined as negatively correlated with redundancy. That equivalence is a definitional framing, not an empirical prediction, and the paper's actual claims of improved accuracy are evaluated against external methods on Cityscapes, ADE20K, COCO-stuff-10K, ImageNet, and COCO2017. The MEWCP background and the exact deletion rule for b = C_l - 1 are credited to independent prior work (Späth 1985; Hosseinian et al. 2017), not to the authors' own previous results. The ablation tables (Appendix C) choose the JS metric and full-resolution features based on Cityscapes validation mIoU, which is a test-benchmark selection concern, but it does not make the headline comparison forced by construction, and the method transfers across datasets. The discrepancy between the unpruned baseline of 81.6 in Table 1 and 79.3 in Table 7 and Appendix C is a reproducibility and correctness risk that should be reconciled, but it is not an instance of circular reasoning.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests mainly on the heuristic quality of the greedy solver and the fidelity of EMA-collected redundancy. No new physical entities are introduced. The method inherits the usual assumptions of data-driven pruning (feature statistics reflect importance) and adds a mild selection-bias component because the metric was chosen on the benchmark used for reporting.

free parameters (3)
  • EMA decay alpha = 0.99
    Decay factor for accumulating edge weights in Eq. 4; fixed without sensitivity analysis.
  • T_step (number of progressive pruning steps) = 1 (ADE20K) or 2 (Cityscapes, COCO stuff)
    Number of prune-finetune cycles in Algorithm 2, chosen per dataset without systematic ablation.
  • Redundancy metric choice = JS divergence
    Selected among dot product, dice score, KL divergence, and JS divergence based on Cityscapes validation mIoU (Table 9).
assumptions (4)
  • domain assumption The pairwise Jensen-Shannon divergence of normalized feature maps is a valid proxy for information redundancy between channels.
    Used to define the edge weights a_ij in Eq. 4; no theoretical guarantee links JS divergence to redundancy relevant for segmentation accuracy.
  • domain assumption EMA of redundancy computed during pre-training reflects the final network's redundancy.
    Algorithm 2 accumulates a_ij during training; if feature statistics shift, the accumulated weights may be stale.
  • ad hoc to paper The greedy deletion of the vertex with minimal summed edge weight yields a near-optimal solution to MEWCP in practice.
    EHGP is justified only empirically via Table 6; no approximation guarantee is provided.
  • standard math MEWCP is NP-hard.
    Cited from Hosseinian et al. 2017; standard complexity result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation." pith.science (2026). https://pith.science/paper/LHIG4B7G

@misc{pith2026241212672,
  author       = {Pith},
  title        = {Pith review of: Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LHIG4B7G}},
  note         = {Machine review of arXiv:2412.12672}
}
read the original abstract

In recent years, semantic segmentation has flourished in various applications. However, the high computational cost remains a significant challenge that hinders its further adoption. The filter pruning method for structured network slimming offers a direct and effective solution for the reduction of segmentation networks. Nevertheless, we argue that most existing pruning methods, originally designed for image classification, overlook the fact that segmentation is a location-sensitive task, which consequently leads to their suboptimal performance when applied to segmentation networks. To address this issue, this paper proposes a novel approach, denoted as Spatial-aware Information Redundancy Filter Pruning~(SIRFP), which aims to reduce feature redundancy between channels. First, we formulate the pruning process as a maximum edge weight clique problem~(MEWCP) in graph theory, thereby minimizing the redundancy among the remaining features after pruning. Within this framework, we introduce a spatial-aware redundancy metric based on feature maps, thus endowing the pruning process with location sensitivity to better adapt to pruning segmentation networks. Additionally, based on the MEWCP, we propose a low computational complexity greedy strategy to solve this NP-hard problem, making it feasible and efficient for structured pruning. To validate the effectiveness of our method, we conducted extensive comparative experiments on various challenging datasets. The results demonstrate the superior performance of SIRFP for semantic segmentation tasks.

Figures

Figures reproduced from arXiv: 2412.12672 by the authors.

Figure 1
Figure 1. Params vs mIOU on Cityscapes dataset. The pro [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Visualization comparison with DCFP (Wang et al. 2024) on Cityscapes validation set with 60% FLOPs reduction of [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Visualization comparison with DCFP (Wang et al. 2024) on ADE20K validation set with 60% FLOPs reduction of [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 50 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Caesar, H.; Uijlings, J.; and Ferrari, V. 2018. COCO-Stuff: Thing and Stuff Classes in Context. In CVPR, 1209--1218

    work page 2018

  4. [4]

    learning-compression

    Carreira-Perpin \'a n, M. A.; and Idelbayev, Y. 2018. “learning-compression” algorithms for neural net pruning. In CVPR, 8532--8541

    work page 2018

  5. [5]

    Chen, L.-C.; Papandreou, G.; Kokkinos, I.; Murphy, K.; and Yuille, A. L. 2014. Semantic image segmentation with deep convolutional nets and fully connected crfs. arXiv preprint arXiv:1412.7062

    arXiv 2014

  6. [6]

    Chen, L.-C.; Papandreou, G.; Kokkinos, I.; Murphy, K.; and Yuille, A. L. 2017 a . Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. TPAMI, 40(4): 834--848

    work page 2017

  7. [7]

    Chen, L.-C.; Papandreou, G.; Schroff, F.; and Adam, H. 2017 b . Rethinking atrous convolution for semantic image segmentation. arXiv preprint arXiv:1706.05587

    arXiv 2017

  8. [8]

    Chen, L.-C.; Zhu, Y.; Papandreou, G.; Schroff, F.; and Adam, H. 2018. Encoder-decoder with atrous separable convolution for semantic image segmentation. In ECCV, 801--818

    work page 2018

Show all 63 references
  1. [9]

    Chen, X.; Zhang, Y.; and Wang, Y. 2022. MTP: multi-task pruning for efficient semantic segmentation networks. In ICME, 1--6

    2022

  2. [10]

    Cordts, M.; Omran, M.; Ramos, S.; Rehfeld, T.; Enzweiler, M.; Benenson, R.; Franke, U.; Roth, S.; and Schiele, B. 2016. The cityscapes dataset for semantic urban scene understanding. In CVPR, 3213--3223

    2016

  3. [11]

    Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; and Fei-Fei, L. 2009. Imagenet: A large-scale hierarchical image database. In CVPR, 248--255

    2009

  4. [12]

    Q.; and Wang, G

    Ding, H.; Jiang, X.; Shuai, B.; Liu, A. Q.; and Wang, G. 2018. Context contrasted feature and gated multi-scale aggregation for scene segmentation. In CVPR, 2393--2402

    2018

  5. [13]

    Ding, X.; Ding, G.; Guo, Y.; Han, J.; and Yan, C. 2019. Approximated oracle filter pruning for destructive cnn width optimization. In ICML, 1607--1616

    2019

  6. [14]

    Dong, X.; and Yang, Y. 2019. Network pruning via transformable architecture search. NeurIPS, 32

    2019

  7. [15]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

    2020 arXiv

  8. [16]

    K.; McKinstry, J

    Esser, S. K.; McKinstry, J. L.; Bablani, D.; Appuswamy, R.; and Modha, D. S. 2019. Learned step size quantization. arXiv preprint arXiv:1902.08153

    2019 arXiv

  9. [17]

    Fan, M.; Lai, S.; Huang, J.; Wei, X.; Chai, Z.; Luo, J.; and Wei, X. 2021. Rethinking bisenet for real-time semantic segmentation. In CVPR, 9716--9725

    2021

  10. [18]

    B.; and Wang, X

    Fang, G.; Ma, X.; Song, M.; Mi, M. B.; and Wang, X. 2023. Depgraph: Towards any structural pruning. In CVPR, 16091--16101

    2023

  11. [19]

    Guo, S.; Wang, Y.; Li, Q.; and Yan, J. 2020. Dmcp: Differentiable markov channel pruning for neural networks. In CVPR, 1539--1547

    2020

  12. [20]

    Han, S.; Mao, H.; and Dally, W. J. 2015. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. arXiv preprint arXiv:1510.00149

    2015 arXiv

  13. [21]

    Han, S.; Pool, J.; Tran, J.; and Dally, W. 2015. Learning both weights and connections for efficient neural network. NeurIPS, 28

    2015

  14. [22]

    He, W.; Wu, M.; Liang, M.; and Lam, S.-K. 2021. Cap: Context-aware pruning for semantic segmentation. In WACV, 960--969

    2021

  15. [23]

    He, Y.; Kang, G.; Dong, X.; Fu, Y.; and Yang, Y. 2018. Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks. In IJCAI, 2234--2240

    2018

  16. [24]

    He, Y.; Liu, P.; Wang, Z.; Hu, Z.; and Yang, Y. 2019. Filter pruning via geometric median for deep convolutional neural networks acceleration. In CVPR, 4340--4349

    2019

  17. [25]

    B.; and Butenko, S

    Hosseinian, S.; Fontes, D. B.; and Butenko, S. 2020. A lagrangian bound on the clique number and an exact algorithm for the maximum edge weight clique problem. INFORMS Journal on Computing, 32(3): 747--762

    2020

  18. [26]

    B.; Butenko, S.; Nardelli, M

    Hosseinian, S.; Fontes, D. B.; Butenko, S.; Nardelli, M. B.; Fornari, M.; and Curtarolo, S. 2017. The maximum edge weight clique problem: formulations and solution approaches. Optimization Methods and Applications: In Honor of Ivan V. Sergienko's 80th Birthday, 217--237

    2017

  19. [27]

    Hou, Z.; Qin, M.; Sun, F.; Ma, X.; Yuan, K.; Xu, Y.; Chen, Y.-K.; Jin, R.; Xie, Y.; and Kung, S.-Y. 2022. Chex: Channel exploration for cnn model compression. In CVPR, 12287--12298

    2022

  20. [28]

    Huang, Z.; and Wang, N. 2018. Data-driven sparse structure selection for deep neural networks. In ECCV, 304--320

    2018

  21. [29]

    Ioffe, S.; and Szegedy, C. 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 448--456

    2015

  22. [30]

    Li, H.; Kadav, A.; Durdanovic, I.; Samet, H.; and Graf, H. 2016. Pruning Filters for Efficient ConvNets. arXiv preprint arXiv: 1608.08710

    2016 arXiv

  23. [31]

    V.; and Timofte, R

    Li, Y.; Gu, S.; Mayer, C.; Gool, L. V.; and Timofte, R. 2020. Group sparsity: The hinge between filter pruning and decomposition for network compression. In CVPR, 8018--8027

    2020

  24. [32]

    Lin, M.; Ji, R.; Wang, Y.; Zhang, Y.; Zhang, B.; Tian, Y.; and Shao, L. 2020 a . Hrank: Filter pruning using high-rank feature map. In CVPR, 1529--1538

    2020

  25. [33]

    Lin, T.; Stich, S.; Barba, L.; Dmitriev, D.; and Jaggi, M. 2020 b . Dynamic Model Pruning with Feedback. ICLR

    2020

  26. [34]

    Lin, T.-Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Doll \'a r, P.; and Zitnick, C. L. 2014. Microsoft coco: Common objects in context. In ECCV, 740--755

    2014

  27. [35]

    Liu, L.; Zhang, S.; Kuang, Z.; Zhou, A.; Xue, J.-H.; Wang, X.; Chen, Y.; Yang, W.; Liao, Q.; and Zhang, W. 2021 a . Group fisher pruning for practical network compression. In ICML, 7021--7032

    2021

  28. [36]

    Liu, Z.; Li, J.; Shen, Z.; Huang, G.; Yan, S.; and Zhang, C. 2017. Learning efficient convolutional networks through network slimming. In CVPR, 2736--2744

    2017

  29. [37]

    Liu, Z.; Lin, Y.; Cao, Y.; Hu, H.; Wei, Y.; Zhang, Z.; Lin, S.; and Guo, B. 2021 b . Swin transformer: Hierarchical vision transformer using shifted windows. In ICCV, 10012--10022

    2021

  30. [38]

    Long, J.; Shelhamer, E.; and Darrell, T. 2015. Fully convolutional networks for semantic segmentation. In CVPR, 3431--3440

    2015

  31. [39]

    Luo, J.-H.; Wu, J.; and Lin, W. 2017. Thinet: A filter level pruning method for deep neural network compression. In CVPR, 5058--5066

    2017

  32. [40]

    M.; and De Souza, C

    Macambira, E. M.; and De Souza, C. C. 2000. The edge-weighted clique problem: valid inequalities, facets and polyhedral computations. European Journal of Operational Research, 123(2): 346--371

    2000

  33. [41]

    Molchanov, P.; Mallya, A.; Tyree, S.; Frosio, I.; and Kautz, J. 2019. Importance estimation for neural network pruning. In CVPR, 11264--11272

    2019

  34. [42]

    Ning, X.; Zhao, T.; Li, W.; Lei, P.; Wang, Y.; and Yang, H. 2020. DSA: More Efficient Budgeted Pruning via Differentiable Sparsity Allocation. In ECCV, 592--607

    2020

  35. [43]

    Oh, J.; Kim, H.; Baik, S.; Hong, C.; and Lee, K. M. 2022. Batch normalization tells you which filter is important. In WACV, 2645--2654

    2022

  36. [44]

    Pan, H.; Hong, Y.; Sun, W.; and Jia, Y. 2022. Deep dual-resolution networks for real-time and accurate semantic segmentation of traffic scenes. T-ITS, 24(3): 3448--3460

    2022

  37. [45]

    Pang, Y.; Li, Y.; Shen, J.; and Shao, L. 2019. Towards bridging semantic gap to improve semantic segmentation. In CVPR, 4230--4239

    2019

  38. [46]

    Peng, H.; Wu, J.; Chen, S.; and Huang, J. 2019. Collaborative channel pruning for deep networks. In ICML, 5113--5122

    2019

  39. [47]

    Pullan, W. 2008. Approximating the maximum vertex/edge weighted clique using local search. Journal of Heuristics, 14: 117--134

    2008

  40. [48]

    Ronneberger, O.; Fischer, P.; and Brox, T. 2015. U-net: Convolutional networks for biomedical image segmentation. In Proc. Int. Conf. Med. Image Comput. Comput.-Assist. Intervent., 234--241

    2015

  41. [49]

    a th, H. 1985. Heuristically determining cliques of given cardinality and with minimal cost within weighted complete graphs. Zeitschrift f \

    Sp \"a th, H. 1985. Heuristically determining cliques of given cardinality and with minimal cost within weighted complete graphs. Zeitschrift f \"u r Operations Research , 29: 125--131

    1985

  42. [50]

    Su, X.; You, S.; Huang, T.; Wang, F.; Qian, C.; Zhang, C.; and Xu, C. 2021. Locally Free Weight Sharing for Network Width Search. ICLR

    2021

  43. [51]

    Sui, Y.; Yin, M.; Xie, Y.; Phan, H.; Aliari Zonouz, S.; and Yuan, B. 2021. Chip: Channel independence-based pruning for compact neural networks. NeurIPS, 34: 24604--24616

    2021

  44. [52]

    Wan, Q.; Huang, Z.; Lu, J.; Yu, G.; and Zhang, L. 2023. Seaformer: Squeeze-enhanced axial transformer for mobile semantic segmentation. arXiv preprint arXiv:2301.13156

    2023 arXiv

  45. [53]

    Wang, H.; Qin, C.; Zhang, Y.; and Fu, Y. 2020. Neural pruning via growing regularization. arXiv preprint arXiv:2012.09243

    2020 arXiv

  46. [54]

    Wang, J.; Gou, C.; Wu, Q.; Feng, H.; Han, J.; Ding, E.; and Wang, J. 2022. RTFormer: Efficient design for real-time semantic segmentation with transformer. NeurIPS, 35: 7423--7436

    2022

  47. [55]

    Wang, Z.; Xie, H.; Wang, Y.; Xu, H.; and Jin, G. 2024. DCFP: Distribution Calibrated Filter Pruning for Lightweight and Accurate Long-tail Semantic Segmentation. TCSVT, 34(7): 6063--6076

    2024

  48. [56]

    Xu, Y.; Wang, Y.; Zhou, A.; Lin, W.; and Xiong, H. 2018. Deep neural network compression with single and multiple level quantization. In AAAI, volume 32

    2018

  49. [57]

    Yang, Z.; Wang, Y.; Liu, C.; Chen, H.; Xu, C.; Shi, B.; Xu, C.; and Xu, C. 2019. Legonet: Efficient convolutional neural networks with lego filters. In ICML, 7005--7014

    2019

  50. [58]

    Yu, C.; Gao, C.; Wang, J.; Yu, G.; Shen, C.; and Sang, N. 2021. Bisenet v2: Bilateral network with guided aggregation for real-time semantic segmentation. IJCV, 129: 3051--3068

    2021

  51. [59]

    Yu, C.; Wang, J.; Peng, C.; Gao, C.; Yu, G.; and Sang, N. 2018. Bisenet: Bilateral segmentation network for real-time semantic segmentation. In ECCV, 325--341

    2018

  52. [60]

    Yu, X.; Liu, T.; Wang, X.; and Tao, D. 2017. On compressing deep models by low rank and sparse decomposition. In CVPR, 7370--7379

    2017

  53. [61]

    Zhao, H.; Shi, J.; Qi, X.; Wang, X.; and Jia, J. 2017. Pyramid scene parsing network. In CVPR, 2881--2890

    2017

  54. [62]

    Zhou, B.; Zhao, H.; Puig, X.; Fidler, S.; Barriuso, A.; and Torralba, A. 2017. Scene parsing through ade20k dataset. In CVPR, 633--641

    2017

  55. [63]

    Zhuang, Z.; Tan, M.; Zhuang, B.; Liu, J.; Guo, Y.; Wu, Q.; Huang, J.; and Zhu, J. 2018. Discrimination-aware channel pruning for deep neural networks. NeurIPS, 31

    2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.