Architecture-aware Network Pruning for Vision Quality Applications

Cheng-Ming Chiang; Han-Lin Li; Wei-Shiang Lin; Wei-Ting Wang; Yi-Min Tsai

REVIEW 3 major objections 4 minor 31 references

Architecture-aware Network Pruning for Vision Quality Applications

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper claims that architecture-aware output-channel pruning cuts low-light imaging compute by 58% and super-resolution compute by 37% with no measured quality drop.

desk verdict The MAC-reduction heuristics are genuinely useful, but the 'without quality drop' claim is weakened by the validation-guided stopping criterion; the paper deserves a serious referee but needs test-set evidence. read the letter →

arxiv 1908.02125 v1 pith:SWLADK2W submitted 2019-08-05 eess.IV cs.CVcs.LG

Wei-Ting Wang , Han-Lin Li , Wei-Shiang Lin , Cheng-Ming Chiang , Yi-Min Tsai This is my paper · ORCID

classification eess.IVcs.CVcs.LG

keywords PruningVisionQualityNetworkArchitectureLow-lightimagingSuper-resolutionMultiply-AccumulatereductionMemorybandwidthOutput-channel

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that convolutional networks for vision quality tasks, which are extremely expensive because they process large images, can be pruned aggressively without sacrificing output quality. The proposed method removes whole output channels whose maximum absolute weight is below a layer-specific adaptive threshold, then iteratively retrains and checks PSNR and SSIM before continuing. This architecture-aware pruning preserves network depth, steers pruning toward layers with higher computation per weight, and handles residual-block connections by pruning grouped channels together. On SID low-light imaging and EDSR super-resolution, the paper reports 58% and 37% Multiply-Accumulate reductions with no PSNR or SSIM drop, plus 39% and 20% memory-bandwidth reductions for convolutional layers. These savings matter because they could bring high-resolution quality-enhancement networks closer to mobile-device deployment without specialized hardware.

What carries the argument

The engine is a per-layer pruning threshold, $T_l = T_b (1 - S_l) R_l$, where $T_b$ is a global base threshold, $S_l$ is the layer's already-pruned output-channel ratio, and $R_l = \log_{10}(M_l/W_l)$ is the layer's MAC-per-weight efficiency. Output channels whose maximum absolute weight falls below $T_l$ are removed, and the corresponding input channels of the following layer are removed too. Raising the threshold for high-$R_l$ and lightly-pruned layers redirects pruning toward layers that save the most computation per removed weight, and grouped channel pruning keeps residual-block element-wise additions valid. Preserving a minimum number of output channels per layer prevents whole layers from disappearing, which the paper reports as necessary to avoid unrecoverable quality loss.

What would settle it

Take a trained SID model, apply the full Method-D pruning, and evaluate on unseen low-light scenes from a different camera or much darker exposures: if PSNR drops appreciably or visible color and edge defects appear even when validation PSNR matches the original, the no-quality-drop claim does not generalize beyond the reported validation setting.

Watch

Extended reading notes

Core claim

The paper's central claim is that a pruning algorithm which removes output channels when each channel's maximum absolute weight falls below a layer-specific adaptive threshold, combined with iterative retraining and a PSNR/SSIM stop criterion, can substantially reduce computation and memory bandwidth in vision quality networks without lowering quality. On SID, multiply-accumulate operations drop from 560G to 236G (58%) with PSNR 28.55 versus 28.54 and SSIM 0.768 versus 0.767. On EDSR, MAC drops from 1428G to 897G (37%) with PSNR and SSIM unchanged at 34.42 and 0.942. Convolutional-layer memory bandwidth drops 39% on SID and 20% on EDSR. The paper also presents example images that look indistinguishable from the unpruned outputs, although the formal guarantee is stated in terms of PSNR and SSIM.

Load-bearing premise

The load-bearing premise is that any output channel whose maximum absolute weight is below the threshold is effectively useless, so removing it cannot destroy information that retraining cannot restore.

Editorial extensions

If this is right

On SID, the full architecture-aware method cuts MAC from 560G to 236G while holding validation PSNR at 28.55 versus 28.54 and SSIM at 0.768 versus 0.767.
On EDSR, MAC drops from 1428G to 897G with PSNR and SSIM exactly matching the original at 34.42 and 0.942.
Convolutional-layer memory bandwidth is reduced by 39% on SID and 20% on EDSR, lowering the memory-bound portion of inference.
Threshold-only pruning without the depth constraint can remove an entire layer and produce quality degradation that retraining does not recover, so architecture awareness is a necessary component.
Weight sparsity alone is not a reliable proxy for compute savings; steering pruning by MAC-per-weight is what converts removed weights into actual Multiply-Accumulate reduction.

Reading between the lines

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

Because the threshold formula depends only on maximum absolute weight, MAC count, and output-channel counts, the same recipe should port to other quality enhancement networks such as deblurring, HDR reconstruction, or denoising; running the method on one of those tasks would test the generality cheaply.
The paper's own example shows a 0.09 PSNR drop can produce visible defects, so the phrase "without quality drop" should be read as metric-bound; a perceptual no-regression check on artifact-prone low-light scenes would test whether the guarantee extends beyond PSNR and SSIM.
Since the reported bandwidth savings come mainly from fewer activations, combining this pruning with activation compression or lower-precision arithmetic could compound the memory gains beyond the reported 20-39%.

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.

Desk Editor's Note

read the letter

The headline result—58% MAC reduction on SID and 37% on EDSR without quality drop—should be read with a grain of salt. The 'without quality drop' is engineered: Algorithm 1 retrains until the validation PSNR/SSIM exceeds a target Q, and the same validation numbers are then reported in Table 1. So those rows are a satisfied constraint, not an independent measurement. The MAC reductions themselves are real arithmetic; whether they generalize is untested because no test set, no error bars, and no code are provided.

What is genuinely new and useful is the architecture-aware layer prioritization. The R_l = log10(MAC/weights) heuristic is a sensible way to focus pruning on layers that actually save computation, and the balanced pruned-output-channel rule for residual blocks is a practical fix for a real headache in channel pruning. The paper explains clearly why keep-layer-depth matters for SID's U-Net and why residual blocks need grouped pruning. The ablation across Methods A-D is decent and shows each heuristic moving the needle for SID. The EDSR result, where Methods A-C are identical and only D helps, is an honest demonstration that the extra heuristics pay off only on shortcut-connected layers.

The soft spots besides the validation circularity: no comparison with any prior pruning method on these tasks (e.g., [12], [17]), so the 58%/37% claims are not contextualized against a baseline. No latency is measured, despite the implied speed-up. The threshold schedule (T_b, T_i, S_i) and retraining hyperparameters are underspecified, which makes reproduction hard. Single-run numbers on small validation sets (93 images for SID, 100 for DIV2K) are noisy.

The core ideas are worth checking. The circularity is load-bearing for the quality claim but not for the MAC-reduction claim, which stands on its own if the structural heuristics are valid. I would send this to peer review, but I would require a held-out test evaluation, error bars or multiple restarts, code or detailed hyperparameters, and at least one comparison with a standard channel-pruning method. The audience is researchers and engineers who need to shrink quality-oriented networks for mobile deployment; they should read it for the heuristics, not for the headline.

For my own work: I wouldn't cite the numbers, but I might cite the residual-balancing rule in a related-work paragraph. Bring it to reading group? Maybe—it is a compact example of a metric being used as a stopping criterion, which is a nice discussion point.

Referee Report

3 major / 4 minor

Summary. The manuscript proposes an iterative output-channel pruning algorithm for vision quality CNNs. It uses an adaptive magnitude threshold modulated by a layer's MAC-to-weight ratio and its current output-channel pruning ratio, preserves a minimum number of output channels per layer to keep layer depth, and balances pruned channels across residual and elementwise-ADD blocks. The algorithm alternates pruning with retraining and stops when a quality metric (PSNR/SSIM) target or a maximum step budget is reached. Experiments on SID (low-light imaging) and EDSR (super-resolution) report MAC reductions of 58% and 37%, respectively, with validation PSNR/SSIM equal to or slightly above the unpruned networks, and bandwidth reductions of 39% and 20%.

Significance. If the reported reductions generalize beyond the validation splits, the paper would offer a practical contribution to deploying quality-sensitive CNN models on resource-constrained devices. The architectural insights are well motivated: preserving layer depth, using the MAC-to-weight ratio to prioritize which layers to prune, and balancing output-channel pruning across residual blocks all address real deployment constraints. The per-layer pruned-channel analysis in Figs. 6 and 7 is informative and goes beyond a single aggregate sparsity number. However, the evidence for the headline 'without quality drop' claim is not currently independent of the stopping mechanism in Algorithm 1, and the evaluation lacks a held-out test set, repeated trials, and any measured latency. The core idea is plausible, but the results as presented do not yet support the strong guarantee stated in the abstract.

major comments (3)

[Sec. 3.2 and Algorithm 1, Table 1] The claim 'without quality drop' is enforced, not independently demonstrated. The inner loop of Algorithm 1 (lines 16-20) retrains until the quality metric on the validation set exceeds Q, and Table 1 then reports the validation PSNR/SSIM of exactly that selected model. If Q is chosen as the original validation score, equality or improvement on that same validation split is a constraint of the search rather than a measured outcome. The paper never specifies the value of Q, nor does it describe any held-out test split or an evaluation protocol in which the stopping criterion is applied to a separate set. To support the abstract claim, the authors should specify Q, report performance on a test set unseen during pruning and stopping, and show the trajectory of the quality metric across outer iterations.
[Sec. 3.1, Eq. (2)] The pruning rule assumes that output channels whose maximum absolute weight is below the adaptive threshold are safe to remove. The paper does not test this assumption against a sensitivity-based baseline or analyze the distribution of pruned weights. For high-resolution, quality-sensitive applications, small-magnitude weights may still encode edge or chroma information, and retraining alone may not recover that capability. The authors should add a comparison with a structured magnitude-based baseline and report the distance between pruned and original feature maps, or otherwise demonstrate that the pruned channels are not quality-critical.
[Table 1 and Sec. 5] All reported numbers are for a single run without error bars or multiple restarts, so the exact equality of PSNR to two decimals (e.g., SID 28.54 and EDSR 34.42) cannot be distinguished from rounding noise, and the word 'without quality drop' is stronger than the evidence. Additionally, the conclusion states that the MAC and bandwidth reductions 'imply reduction on inference latency', but no inference latency is measured anywhere in the paper. The authors should report mean and standard deviation over at least three independent pruning-retraining runs, and either measure latency on a target device or rephrase the latency statement as a potential benefit rather than an implication.

minor comments (4)

[Sec. 4.2, EDSR paragraph] The sentence 'there is no difference among Method-A, Method-B and Method-C because no layer is pruned by Method-A and MAC/weight are identical for all layers' appears to contradict Table 1, which reports that Method-A reduces EDSR MAC to 76% of the original; please clarify this sentence and the intended comparison.
[Fig. 2] The y-axis label reads 'MACs per weight(Log 10)' but the axis limits and the mapping of layer indices to SID and EDSR are not explained; please state that the figure plots Eq. 1 and clarify which layer indices correspond to the 'top and bottom layers' of SID mentioned in the text.
[Sec. 3.1.3 and Eq. 2] The balancing guidance is described for residual blocks with element-wise ADD, but Algorithm 1 applies Eq. 2 to every layer; please state explicitly how S_l is computed for a non-residual layer and how grouped output channels are synchronized when the following layer's input channels are removed.
[Algorithm 1] The instruction 'jump to line 2' makes the incremental target sparsity S = S_i + total-network-sparsity depend on the current network sparsity, but the term 'total-network-sparsity' is not defined as an initialization value; please define it and report how many outer iterations were executed for each result in Table 1.

Circularity Check

1 steps flagged · score 6.0 of 10

The 'without quality drop' claim is enforced by Algorithm 1's validation-quality stopping rule, so the reported validation PSNR/SSIM are not an independent prediction; the MAC-reduction figures are independently grounded.

fitted input called prediction [Sec. 3.2, Algorithm 1 (lines 16-20); Table 1]
"To maintain quality metric (PSNR and SSIM) while maximizing pruned MAC, our algorithm prunes and retrains network iteratively. The iteration terminates when either the target quality-metric criteria or maximum training steps is reached."

Algorithm 1's inner loop retrains until 'Qt > Q or g >= G', where Qt is the quality metric evaluated on the validation set and Q is the target quality. The same validation PSNR/SSIM values are then reported in Table 1 as evidence of 'no quality drop'. If Q is set to the original model's validation PSNR/SSIM, the final reported validation values are forced to be at least Q by the stopping condition (unless the step limit is hit). Thus the 'without quality drop' result is a satisfied constraint, not an independently measured or predicted outcome. The paper also describes no held-out test set or repeated runs, so the validation numbers cannot validate generalization beyond the split used for stopping.

full rationale

The paper's MAC and bandwidth reductions are computed from the pruned architecture relative to the original model and are independent of the validation stopping rule; those efficiency numbers are not circular. However, the headline 'without quality drop' claim reduces, on the validation set, to Algorithm 1's explicit stopping criterion: the algorithm prunes and retrains until Qt > Q, then Table 1 reports the resulting validation PSNR/SSIM. Since the reported quality numbers are the same quantities used as the termination test, the absence of quality drop is enforced by construction rather than demonstrated as a prediction. The paper further frames quality metrics as 'a reference judgement for termination of pruning procedure,' confirming that the quality outcome is an input condition, not an independent finding. This is a partial circularity: one part of the central claim is self-confirming on the reported split, while the MAC/BW results retain independent content. No self-citation chain or imported uniqueness theorem is involved.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The algorithm rests on several engineering assumptions. The pruning criterion is a heuristic with no theoretical guarantee. Quality is defined operationally as validation PSNR/SSIM and used as the stopping condition. Retraining is assumed to restore quality, and residual connectivity is assumed to require grouped same-index channel removal. The MAC-efficiency priority R_l is introduced ad hoc. No new physical or mathematical entities are postulated. Free hyperparameters are numerous and mostly unreported, so the ledger is dominated by engineering choices rather than physical constants.

free parameters (7)

Threshold base T_b = not reported
Used in Eq. 2 to set each layer's pruning threshold and incremented by T_i until target sparsity. The value is hand-chosen and not disclosed, and it directly controls how many channels are pruned.
Threshold increment T_i = not reported
Controls the speed of iterative pruning in Algorithm 1; affects the final pruned structure and is never given a value.
Target sparsity increment S_i = not reported
Used in Algorithm 1 to set how much sparsity is added per iteration; no value is provided.
Target quality Q = implicitly original validation PSNR/SSIM
Algorithm 1 stops when validation quality exceeds Q; this is the quality bar that defines 'no quality drop'.
Maximum training steps G = not reported
Termination condition for retraining in Algorithm 1; not provided.
Minimum output channels per layer = not reported
Introduced in Sec. 3.1.1 to preserve network depth; the threshold or minimum count is not specified and affects results.
Retraining hyperparameters (optimizer, learning rate, epochs) = not reported
Retraining is central to quality recovery, but no schedule is given, so exact reproduction is impossible.

assumptions (6)

domain assumption Small-magnitude output channels are unimportant.
Core magnitude pruning criterion in Sec. 3.1: channels whose maximum absolute weight is below threshold are pruned. This is assumed to preserve quality after retraining.
domain assumption Validation PSNR/SSIM capture subjective quality.
Sec. 3.2 and Fig. 1 treat PSNR/SSIM as the quality metric for stopping; zero drop on validation is equated with no quality drop.
domain assumption Retraining restores quality after pruning.
Algorithm 1 assumes pruned networks can be retrained to meet the target quality within G steps.
domain assumption Residual ADD requires grouped pruning of same-index channels.
Sec. 3.1.3 and Fig. 3: element-wise ADD forces channels with the same index to be pruned together; this is a structural assumption about EDSR's residual blocks.
ad hoc to paper R_l=log10(M_l/W_l) is a useful priority metric.
Sec. 3.1.2 introduces R_l as a MAC-efficiency indicator without derivation or validation; used to bias pruning toward high-MAC layers.
domain assumption Output-channel-wise coarse pruning is preferable to fine-grained pruning.
Sec. 2 relies on [18] and [19] to justify coarse-grained pruning for hardware efficiency.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Architecture-aware Network Pruning for Vision Quality Applications." pith.science (2026). https://pith.science/paper/SWLADK2W

@misc{pith2026190802125,
  author       = {Pith},
  title        = {Pith review of: Architecture-aware Network Pruning for Vision Quality Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SWLADK2W}},
  note         = {Machine review of arXiv:1908.02125}
}

read the original abstract

Convolutional neural network (CNN) delivers impressive achievements in computer vision and machine learning field. However, CNN incurs high computational complexity, especially for vision quality applications because of large image resolution. In this paper, we propose an iterative architecture-aware pruning algorithm with adaptive magnitude threshold while cooperating with quality-metric measurement simultaneously. We show the performance improvement applied on vision quality applications and provide comprehensive analysis with flexible pruning configuration. With the proposed method, the Multiply-Accumulate (MAC) of state-of-the-art low-light imaging (SID) and super-resolution (EDSR) are reduced by 58% and 37% without quality drop, respectively. The memory bandwidth (BW) requirements of convolutional layer can be also reduced by 20% to 40%.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 14 canonical work pages

[12]

Enhanced deep residual net- works for single image super-resolution,

Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee, “Enhanced deep residual net- works for single image super-resolution,” CoRR, vol. abs/1707.02921, 2017

arXiv 2017
[17]

A new pruning heuristic based on variance analysis of sensitivity information,

A. P. Engelbrecht, “A new pruning heuristic based on variance analysis of sensitivity information,” Trans. Neur. Netw., vol. 12, no. 6, pp. 1386–1399, Nov. 2001

work page 2001
[1]

Vision perception tasks including image classiﬁcation, object detection and seman- tic segmentation are comprehensively investigated and asso- ciated with CNN

INTRODUCTION CNN is adopted as an essential ingredients in computer vision and machine learning areas [1, 2, 3]. Vision perception tasks including image classiﬁcation, object detection and seman- tic segmentation are comprehensively investigated and asso- ciated with CNN. Even in image processing ﬁeld such as su- per resolution, high dynamic range imaging...

work page 2019
[2]

To answer which weight should be pruned, some works add evaluation functions to loss function, such as group lasso

RELATED WORKS Network pruning has been widely explored in existing litera- tures. To answer which weight should be pruned, some works add evaluation functions to loss function, such as group lasso

work page
[3]

Architecture-aware Pruning An output channel is pruned if its maximum absolute weight value is less than magnitude threshold

PROPOSED METHOD 3.1. Architecture-aware Pruning An output channel is pruned if its maximum absolute weight value is less than magnitude threshold. For convolutional layer, the weight kernel has tensor shapei ×o ×k ×k, where i is the number of input channels, o is the number of output channels and k is kernel size. Output-channel-wise pruning removes the w...

work page
[4]

Experiment Setup We generally investigate both SID for low-light photography and EDSR (baseline network, ×2) for super resolution

EXPERIMENTAL RESULT 4.1. Experiment Setup We generally investigate both SID for low-light photography and EDSR (baseline network, ×2) for super resolution. SID uses its own dataset [4] and EDSR adopts DIV2K dataset (a) (b) (c) (d) Fig. 4. SID results of our method compared to original (with- out pruning). (a)(c) Original (PSNR: 28.54, SSIM: 0.767). (b)(d)...

work page 1915
[5]

The MAC of SID and EDSR are reduced by 58% and 37%, respec- tively

CONCLUSION To minimize computation complexity without quality drop on vision quality applications, our architecture-aware prun- ing is optimized for pruning more for complexity metric (e.g., MAC) on SID and shortcut-connected layers on EDSR. The MAC of SID and EDSR are reduced by 58% and 37%, respec- tively. Memory bandwidth is also reduced without degrad...

work page
[6]

Mobilenets: Efﬁcient convolutional neural networks for mobile vision appli- cations,

Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam, “Mobilenets: Efﬁcient convolutional neural networks for mobile vision appli- cations,” CoRR, vol. abs/1704.04861, 2017

arXiv 2017

Show all 31 references

[7]

Imagenet classiﬁcation with deep convolutional neural networks,

Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hin- ton, “Imagenet classiﬁcation with deep convolutional neural networks,” in Advances in Neural Information Processing Systems 25 , F. Pereira, C. J. C. Burges, L. Bottou, and K. Q. Weinberger, Eds., pp. 1097–1105. Curran Assoc...

2012
[8]

However, it is difﬁcult to ﬁnd a proper ratio between additional pruning-related loss and original loss

and MAC regularization [9]. However, it is difﬁcult to ﬁnd a proper ratio between additional pruning-related loss and original loss. Others works create evaluation functions, including sensitivity [10, 11] and weight magnitude [12]. The sensitivity method computes the impact o...

1908 arXiv
[9]

Rich feature hierarchies for accurate ob- ject detection and semantic segmentation,

Ross B. Girshick, Jeff Donahue, Trevor Darrell, and Ji- tendra Malik, “Rich feature hierarchies for accurate ob- ject detection and semantic segmentation,” CoRR, vol. abs/1311.2524, 2013

2013 arXiv
[10]

Fully convolutional networks for semantic segmenta- tion,

Jonathan Long, Evan Shelhamer, and Trevor Darrell, “Fully convolutional networks for semantic segmenta- tion,” CoRR, vol. abs/1411.4038, 2014

2014 arXiv
[11]

Learning to see in the dark,

Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun, “Learning to see in the dark,” CoRR, vol. abs/1805.01934, 2018

2018 arXiv
[13]

Pruning algorithms-a survey,

R. Reed, “Pruning algorithms-a survey,” IEEE Trans- actions on Neural Networks, vol. 4, no. 5, pp. 740–747, Sep. 1993

1993
[14]

Learning structured sparsity in deep neural networks,

Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li, “Learning structured sparsity in deep neural networks,” CoRR, vol. abs/1608.03665, 2016

2016 arXiv
[15]

Morphnet: Fast & sim- ple resource-constrained structure learning of deep net- works,

Ariel Gordon, Elad Eban, Oﬁr Nachum, Bo Chen, Tien- Ju Yang, and Edward Choi, “Morphnet: Fast & sim- ple resource-constrained structure learning of deep net- works,” CoRR, vol. abs/1711.06798, 2017

2017 arXiv
[16]

Op- timal brain damage,

Yann Le Cun, John S. Denker, and Sara A. Solla, “Op- timal brain damage,” in Proceedings of the 2Nd In- ternational Conference on Neural Information Process- ing Systems, Cambridge, MA, USA, 1989, NIPS’89, pp. 598–605, MIT Press

1989
[18]

Learning both weights and connections for efﬁcient neural networks,

Song Han, Jeff Pool, John Tran, and William J. Dally, “Learning both weights and connections for efﬁcient neural networks,” CoRR, vol. abs/1506.02626, 2015

2015 arXiv
[19]

Training sparse neural net- works,

Suraj Srinivas, Akshayvarun Subramanya, and R. Venkatesh Babu, “Training sparse neural net- works,” CoRR, vol. abs/1611.06694, 2016

2016 arXiv
[20]

Compact deep con- volutional neural networks with coarse pruning,

Sajid Anwar and Wonyong Sung, “Compact deep con- volutional neural networks with coarse pruning,”CoRR, vol. abs/1610.09639, 2016

2016 arXiv
[21]

Pruning convolutional neural net- works for resource efﬁcient transfer learning,

Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz, “Pruning convolutional neural net- works for resource efﬁcient transfer learning,” CoRR, vol. abs/1611.06440, 2016

2016 arXiv
[22]

Channel-level acceleration of deep face representations,

A. Polyak and L. Wolf, “Channel-level acceleration of deep face representations,” IEEE Access , vol. 3, pp. 2163–2175, 2015

2015
[23]

Pruning ﬁlters for efﬁcient con- vnets,

Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf, “Pruning ﬁlters for efﬁcient con- vnets,” CoRR, vol. abs/1608.08710, 2016

2016 arXiv
[24]

ADC: automated deep com- pression and acceleration with reinforcement learning,

Yihui He and Song Han, “ADC: automated deep com- pression and acceleration with reinforcement learning,” CoRR, vol. abs/1802.03494, 2018

2018 arXiv
[25]

Exploring the reg- ularity of sparse structure in convolutional neural net- works,

Huizi Mao, Song Han, Jeff Pool, Wenshuo Li, Xingyu Liu, Yu Wang, and William J. Dally, “Exploring the reg- ularity of sparse structure in convolutional neural net- works,” CoRR, vol. abs/1705.08922, 2017

2017 arXiv
[26]

Designing energy-efﬁcient convolutional neural net- works using energy-aware pruning,

Tien-Ju Yang, Yu-Hsin Chen, and Vivienne Sze, “Designing energy-efﬁcient convolutional neural net- works using energy-aware pruning,” CoRR, vol. abs/1611.05128, 2016

2016 arXiv
[27]

Network trimming: A data-driven neuron pruning approach towards efﬁcient deep architectures,

Hengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang, “Network trimming: A data-driven neuron pruning approach towards efﬁcient deep architectures,” CoRR, vol. abs/1607.03250, 2016

2016 arXiv
[28]

An it- erative pruning algorithm for feedforward neural net- works,

G. Castellano, A. M. Fanelli, and M. Pelillo, “An it- erative pruning algorithm for feedforward neural net- works,” IEEE Transactions on Neural Networks , vol. 8, no. 3, pp. 519–531, May 1997

1997
[29]

U-net: Convolutional networks for biomedical image segmentation,

Olaf Ronneberger, Philipp Fischer, and Thomas Brox, “U-net: Convolutional networks for biomedical image segmentation,” CoRR, vol. abs/1505.04597, 2015

2015 arXiv
[30]

Deep residual learning for image recognition,

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Deep residual learning for image recognition,” CoRR, vol. abs/1512.03385, 2015

2015 arXiv
[31]

Ntire 2017 chal- lenge on single image super-resolution: Dataset and study,

Eirikur Agustsson and Radu Timofte, “Ntire 2017 chal- lenge on single image super-resolution: Dataset and study,” July 2017

2017

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [12]

Enhanced deep residual net- works for single image super-resolution,

Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee, “Enhanced deep residual net- works for single image super-resolution,” CoRR, vol. abs/1707.02921, 2017

arXiv 2017

[2] [17]

A new pruning heuristic based on variance analysis of sensitivity information,

A. P. Engelbrecht, “A new pruning heuristic based on variance analysis of sensitivity information,” Trans. Neur. Netw., vol. 12, no. 6, pp. 1386–1399, Nov. 2001

work page 2001

[3] [1]

Vision perception tasks including image classiﬁcation, object detection and seman- tic segmentation are comprehensively investigated and asso- ciated with CNN

INTRODUCTION CNN is adopted as an essential ingredients in computer vision and machine learning areas [1, 2, 3]. Vision perception tasks including image classiﬁcation, object detection and seman- tic segmentation are comprehensively investigated and asso- ciated with CNN. Even in image processing ﬁeld such as su- per resolution, high dynamic range imaging...

work page 2019

[4] [2]

To answer which weight should be pruned, some works add evaluation functions to loss function, such as group lasso

RELATED WORKS Network pruning has been widely explored in existing litera- tures. To answer which weight should be pruned, some works add evaluation functions to loss function, such as group lasso

work page

[5] [3]

Architecture-aware Pruning An output channel is pruned if its maximum absolute weight value is less than magnitude threshold

PROPOSED METHOD 3.1. Architecture-aware Pruning An output channel is pruned if its maximum absolute weight value is less than magnitude threshold. For convolutional layer, the weight kernel has tensor shapei ×o ×k ×k, where i is the number of input channels, o is the number of output channels and k is kernel size. Output-channel-wise pruning removes the w...

work page

[6] [4]

Experiment Setup We generally investigate both SID for low-light photography and EDSR (baseline network, ×2) for super resolution

EXPERIMENTAL RESULT 4.1. Experiment Setup We generally investigate both SID for low-light photography and EDSR (baseline network, ×2) for super resolution. SID uses its own dataset [4] and EDSR adopts DIV2K dataset (a) (b) (c) (d) Fig. 4. SID results of our method compared to original (with- out pruning). (a)(c) Original (PSNR: 28.54, SSIM: 0.767). (b)(d)...

work page 1915

[7] [5]

The MAC of SID and EDSR are reduced by 58% and 37%, respec- tively

CONCLUSION To minimize computation complexity without quality drop on vision quality applications, our architecture-aware prun- ing is optimized for pruning more for complexity metric (e.g., MAC) on SID and shortcut-connected layers on EDSR. The MAC of SID and EDSR are reduced by 58% and 37%, respec- tively. Memory bandwidth is also reduced without degrad...

work page

[8] [6]

Mobilenets: Efﬁcient convolutional neural networks for mobile vision appli- cations,

Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam, “Mobilenets: Efﬁcient convolutional neural networks for mobile vision appli- cations,” CoRR, vol. abs/1704.04861, 2017

arXiv 2017