SeesawFaceNets: sparse and robust face verification model for mobile platform

Jintao Zhang

REVIEW 3 major objections 6 minor 64 references

SeesawFaceNets: sparse and robust face verification model for mobile platform

T0 review · 3 major / 6 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 This paper claims that SeesawFaceNets, a sparse mobile network built from uneven group convolutions and channel shuffle/share blocks, matches or beats larger face-verification models on public benchmarks while using a fraction of the…

desk verdict Honest engineering report on a lightweight face-verification net; the controlled comparison against a MobileFaceNet reimplementation is the real contribution, but the mobile-efficiency claim rests on MAdds, not measured latency, and the Seesaw block is never isolated from SE/swish/embedding-size changes. read the letter →

arxiv 1908.09124 v3 pith:ITTNNZRE submitted 2019-08-24 cs.CV cs.LG

Jintao Zhang This is my paper · ORCID

classification cs.CVcs.LG

keywords faceverificationmobiledeeplearningSeesawblockunevengroupconvolutionchannelshufflelightweightCNNArcmodelefficiency

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

This paper tries to show that accurate face verification can run on mobile devices without cloud offload by redesigning the inverted bottleneck around the Seesaw block: uneven pointwise group convolutions with channel permute or share operations. The claimed result is that SeesawFaceNets beats MobileFaceNet, the baseline, with 66% of its MAdds, reaches comparable accuracy to MobiFace with 54.2% of its parameters and 31.6% of its MAdds, and that a deeper and wider variant, DW-SeesawFaceNet V2, nearly matches ArcFace on all five validation datasets while using 6.5% of ArcFace's parameters and 4.35% of its MAdds. If correct, this makes privacy-preserving, on-device face verification practical on phones and in poor-connectivity areas. A sympathetic reader would take the paper as an architectural-efficiency claim: the Seesaw block plus squeeze-and-excitation and a 512-dimensional embedding buys large compute savings at small accuracy cost.

What carries the argument

The Seesaw block: a bottleneck residual block whose two pointwise 1x1 convolutions are replaced by uneven group convolutions, with a channel permute/shuffle (Seesaw-shuffle) or a channel-share (Seesaw-share) operation between the two groups so information flows across groups; the paper adds nonlinearity after the second pointwise convolution and wraps the block with squeeze-and-excitation. This block is the mechanism that cuts MAdds because each pointwise convolution is split into unequal groups, so fewer multiply-adds are needed than for a dense 1x1 convolution, while the shuffle/share step prevents the groups from learning isolated features.

What would settle it

Measure the same SeesawFaceNets on a phone or embedded board and compare wall-clock latency and power against MobileFaceNet and ArcFace; if the 146M-MAdds model is not meaningfully faster or if DW-SeesawFaceNet V2's accuracy drops more than a couple of points when the embedding size is reduced to 128 or when squeeze-and-excitation blocks are removed, the central efficiency claim would be falsified.

Watch

Extended reading notes

Core claim

The central discovery is that replacing the inverted residual bottleneck in MobileFaceNet with a modified Seesaw block, consisting of uneven pointwise group convolutions with channel permute/shuffle (Seesaw-shuffle) or channel share (Seesaw-share), plus Swish activation and squeeze-and-excitation, yields a smaller, cheaper network that is more accurate for face verification. In the paper's experiments, Seesaw-shuffleFaceNet reaches 99.70% on LFW and 96.85% on AgeDB-30 with 1.3M parameters and 146M MAdds, outperforming their re-implemented MobileFaceNet (1.2M parameters, 221M MAdds) on every listed dataset. The deeper and wider DW-SeesawFaceNet V2 (4.2M parameters, 526M MAdds) scores 99.80% LFW, 97.24% CFP-FP, 97.55% AgeDB-30, 91.98% CPLFW, and 95.98% CALFW, within 1.2 points of ArcFace (65M parameters, 12.1G MAdds) on every dataset where ArcFace leads, and slightly above ArcFace on CALFW.

Load-bearing premise

The load-bearing premise is that MAdds faithfully predicts real on-device speed and energy; the paper itself notes that channel permute/shuffle and memory-transfer operations cost time and power that MAdds does not count, and no ablation isolates the Seesaw block from the other changes such as embedding size, Swish, squeeze-and-excitation, and network width.

Editorial extensions

If this is right

If the reported numbers hold, a 4.2M-parameter face model can replace a 65M-parameter cloud model on LFW, CFP-FP, AgeDB-30, CPLFW, and CALFW with near-identical accuracy, enabling on-device verification.
SeesawFaceNets demonstrates that the Seesaw block transfers from ImageNet classification to face verification, and that the transfer works with small expansion ratios and short 16-epoch training schedules.
The Seesaw-share block avoids channel permute/shuffle overhead, so high-level framework implementations lose less of the theoretical MAdds saving.
Because the deeper DW-SeesawFaceNet can be trained from scratch with a batch size of 128 and roughly 22GB of GPU memory, such models are within reach of a single commodity GPU rather than a large training cluster.

Reading between the lines

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

Beyond the paper, the accuracy gain over MobileFaceNet should not be attributed to the Seesaw block alone: the paper also changes the embedding size to 512, uses Swish, adds squeeze-and-excitation, and alters network width, so the contribution of the block itself is not isolated.
Beyond the paper, a testable extension is to benchmark Seesaw-shuffle versus Seesaw-share on actual phone hardware, since the paper acknowledges that channel permute and memory-transfer costs are not captured in MAdds and may narrow the gap.
Beyond the paper, the DW-SeesawFaceNet results suggest that depth and width scaling of the Seesaw block yields larger gains on pose and age benchmarks than on LFW, hinting that the shuffle/share mechanism increases feature diversity in a way that mainly helps harder verification conditions.
Beyond the paper, one could test the architecture's robustness on additional benchmarks such as MegaFace or IJB-C, which are not used in the paper, to see whether the near-ArcFace accuracy holds outside the five reported validation sets.

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.

Desk Editor's Note

read the letter

Zhen, quick read of SeesawFaceNets. The useful part is not the architecture itself—Seesaw block is from the author's own Seesaw-Net, and the other pieces (SE, swish, 512-D embedding) are standard. The useful part is the controlled comparison: SeesawFaceNets variants and a reimplemented MobileFaceNet are trained on the same MS1MV2 data with the same schedule and compared directly. On LFW/CFP-FP/AgeDB-30, the Seesaw variants consistently beat their MobileFaceNet reimplementation while using 66% of the MAdds. That is a legit, reproducible-by-reimplementation result and it makes the paper worth a referee's time despite the modest novelty.

But the central headline—'mobile platform' efficiency—is weaker than it looks. Every cost ratio is MAdds, and the authors themselves note in Sections 3.1 and 4.3 that channel permute/shuffle and memory transfer add time and power on mobile. No on-device latency or energy measurement appears anywhere. So the 66%/31.6%/4.35% numbers are theoretical upper bounds on compute savings; real speedups depend on kernel support for uneven group convolutions, which the paper doesn't demonstrate. This is not a minor caveat, because the abstract sells 'computational cost' as if it were device runtime.

The accuracy claims are more solid, with two caveats. Single-run, no error bars—standard for this subfield, but differences of 0.2% on LFW can be noise. More importantly, there is no ablation isolating the Seesaw block from the SE, swish, and 512-dim embedding changes. The reader's suspicion is fair: the gains over MobileFaceNet(*) could come from any of those. An ablation would settle it.

Minor stuff: SeesawFaceNets variants report only three datasets in Table 4, while Section 4.3 says 'all listed'—that overreaches. No code is provided, which hurts reproducibility given the group-convolution details. Citation pattern is fine; self-citation to Seesaw-Net is motivated by the block being reused.

Bottom line: a modest but honest engineering paper with one strong comparison and one over-stated efficiency claim. It deserves full peer review, not desk rejection; it also needs real on-device numbers, an ablation, and error bars before it should be published as a mobile efficiency result. I'd send it out.

Referee Report

3 major / 6 minor

Summary. The paper proposes SeesawFaceNets, a family of lightweight face-verification networks built around the author's previously introduced Seesaw block (uneven pointwise group convolutions with channel shuffle or channel sharing), combined with SE blocks and swish activations. It reports that SeesawFaceNets outperform a reimplemented MobileFaceNet baseline with 66% of the MAdds, achieve accuracy close to MobiFace with about half the parameters and one-third the MAdds, and that a deeper/wider variant (DW-SeesawFaceNet V2) is competitive with ArcFace on five public benchmarks with 6.5% of ArcFace's parameters and 4.35% of its MAdds. The empirical basis is Tables 4 and 5, with training on MS1MV2 and ArcFace loss.

Significance. If the reported numbers are reproducible, the main value is the demonstration that a carefully modified MobileFaceNet-style architecture with Seesaw blocks can reach near-SOTA face-verification accuracy at substantially reduced theoretical MAdds and parameter counts. The paper's direct comparison against its own MobileFaceNet reimplementation using the same training data, embedding size, and batch sizes is a methodological strength relative to many mobile-face papers. The architecture tables are sufficiently detailed to be reimplemented. However, the significance of the efficiency claim is qualified by the lack of on-device measurements and by the absence of ablations isolating the Seesaw block from other modifications.

major comments (3)

[Sections 4.1 and 4.3, Tables 4-5] The central mobile-efficiency claim is supported only by theoretical MAdds, and the authors themselves note that channel permute/shuffle and memory-transfer operations add on-device overhead (Sections 3.1 and 4.3). Since the paper's stated goal is mobile deployment, the 66%/31.6%/4.35% cost ratios should be validated by at least one on-device latency or energy measurement, or the claims should be explicitly limited to theoretical FLOP counts. Without this, the efficiency half of the central claim is not established.
[Sections 3.2 and 5, Table 4] The comparison that motivates the title (SeesawFaceNets vs MobileFaceNet) changes three factors at once: the basic block (Seesaw vs inverted residual), the activation (swish vs presumably PReLU/ReLU in the baseline), and the addition of SE blocks. The paper credits Seesaw blocks for the gains but provides no ablation that isolates the block. A simple control (MobileFaceNet with swish and SE, or SeesawFaceNets without them) is needed to support the attribution.
[Section 5.1, Table 4] The claimed parity with MobiFace is based on a single near-saturated dataset (LFW: 99.65 vs 99.70), with SeesawFaceNets trained on 5.8M images for 16 epochs whereas MobiFace used 3.8M images and 1024 epochs. The abstract's statement that SeesawFaceNets are 'comparable' to MobiFace is stronger than the evidence; either report MobiFace scores on the other four benchmarks or soften the claim.

minor comments (6)

[Throughout] The text contains numerous grammatical errors and typos ('prectical', 'time-comsuming', 'conputational', 'the the') that impede readability; a careful proofreading is needed.
[Section 4.1] The justification for excluding sigmoid/swish MAdds is unclear ('mainstream deep learning hardware will include dedicated processing unit for transcendental functions implement'); state the counting convention explicitly and consistently.
[Section 3.3] The discussion of the linear scaling rule is informal and not tested; clarify whether the learning-rate schedule was tuned separately for each model.
[Tables 1, 2, 6] The architecture tables would benefit from a column indicating MAdds per layer so that the total MAdds can be verified.
[Table 4 footnote] The footnote 'Our implement' should be 'Our implementation' and should state the exact embedding size and loss for all rows.
[Section 5.2] Specify which blocks in Table 6 get the additional skip-connection branch; 'all Inverted bottleneck blocks without residual structure whose filter stride is 2' is ambiguous.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reported accuracy and efficiency results are empirical measurements on public benchmarks, not derived from fitted inputs or self-citation chains.

full rationale

The paper's central claims are empirical: SeesawFaceNets accuracy on LFW, CFP-FP, AgeDB-30, CPLFW, and CALFW, and its parameter/MAdds counts relative to MobileFaceNet and other baselines. These numbers come from actual training runs and arithmetic counting of architecture operations. The Seesaw block is explicitly adopted from the author's prior Seesaw-Net paper [2], and that adoption is a design choice, not a derivation of the current results. No equation or procedure in the paper fits a parameter to the reported accuracies and then re-presents that fit as a prediction; the efficiency ratios (66%, 31.6%, 4.35%) are direct ratios of counted Params/MAdds, not outputs of a model. The paper also does not invoke the prior Seesaw-Net work to forbid alternatives or establish uniqueness; it simply cites it as the origin of the building block. Potential weaknesses, such as absence of on-device latency measurements and lack of an ablation isolating the Seesaw block from SE/swish/embedding-size changes, are experimental validity concerns rather than circularity. Therefore the derivation chain is not circular, and a non-finding is appropriate.

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

The central claim is purely empirical: measured accuracy and theoretical MAdds of a specific network design. It rests on the assumption that the Seesaw block from prior work retains its properties in a new task, that MAdds is a valid efficiency metric, and that standard benchmarks and training recipes are reliable. The hand-chosen architecture hyperparameters (embedding size, depth, width, activation, batch size) are free parameters that influence the reported results.

free parameters (6)

Embedding dimension = 512
Section 3.2 states the embedding size is fixed to 512 instead of 128 in MobileFaceNet because 'better performance could be obtained' with little extra compute. This is a hand-picked hyperparameter that affects accuracy and cost.
Seesaw block variant (shuffle vs share) = shuffle and share
Two variants are proposed; the share variant is claimed to avoid the cost of channel permute/shuffle operations (Section 3.1). The choice of variant changes the reported MAdds and accuracy.
Activation function = swish
Section 3.1 says swish is used because it is 'slightly better' than PReLU for face verification. This is a design choice.
Batch size = 160/192/256
Section 4.2 says batch size is chosen due to limited GPU memory; values differ across models and could affect convergence and final accuracy.
Learning rate schedule = initial LR 0.1, decay at epochs 9, 13, 15
Section 4.2 specifies this schedule for the 16-epoch run, based on the baseline. It affects the trained model's accuracy.
Network depth and width per stage = channels 64-512, RBlocks counts 4,6,2 (Table 1)
The architecture dimensions are hand-designed; no architecture search is performed. These choices determine the parameter count and MAdds.

assumptions (6)

domain assumption The Seesaw block described in [2] is an effective building block and transfers from image classification to face verification.
The paper adopts the block from the author's prior Seesaw-Net without re-deriving or independently validating it; Section 3.1 states 'Seesaw block is introduced in [2]' and it is used as-is with a minor modification.
domain assumption MAdds of convolutional and fully connected layers is a valid measure of computational cost for mobile efficiency, ignoring sigmoid/swish and channel shuffle overhead.
Section 4.1 states 'Here we only count MAdds of the convolutional and fully connected layers, and sigmoid operations not take into account.' All efficiency comparisons rely on this.
domain assumption Published face-verification benchmark numbers (LFW, CFP-FP, AgeDB-30, CPLFW, CALFW) are comparable across papers despite differences in training data, preprocessing, and evaluation details.
The paper compares its results to published numbers for MobiFace, MobileFaceNet, and ArcFace without re-running those baselines under identical protocols.
domain assumption Standard deep learning training practices (SGD with momentum, ArcFace loss with margin 0.5, learning rate decay) are sufficient to train the network from scratch.
Section 4.2 describes the optimizer and loss; the paper does not justify these choices beyond citing the baseline.
domain assumption The 'linear scaling rule' for batch size and learning rate does not materially affect outcomes in the narrow batch-size range used (160-256 vs. 512-1024 in comparisons).
Section 3.3 hypothesizes this without testing; if false, the comparisons to models trained with larger batches are not fully controlled.
domain assumption MS1MV2 is a clean and correctly aligned training set, and MTCNN preprocessing is consistent across experiments.
The paper uses the public MS1MV2 dataset and MTCNN preprocessing (Section 4.2) without additional verification.

how reviews work

0 comments

Cite this review

Pith. "Pith review of SeesawFaceNets: sparse and robust face verification model for mobile platform." pith.science (2026). https://pith.science/paper/ITTNNZRE

@misc{pith2026190809124,
  author       = {Pith},
  title        = {Pith review of: SeesawFaceNets: sparse and robust face verification model for mobile platform},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ITTNNZRE}},
  note         = {Machine review of arXiv:1908.09124}
}

read the original abstract

Deep Convolutional Neural Network (DCNNs) come to be the most widely used solution for most computer vision related tasks, and one of the most important application scenes is face verification. Due to its high-accuracy performance, deep face verification models of which the inference stage occurs on cloud platform through internet plays the key role on most prectical scenes. However, two critical issues exist: First, individual privacy may not be well protected since they have to upload their personal photo and other private information to the online cloud backend. Secondly, either training or inference stage is time-comsuming and the latency may affect customer experience, especially when the internet link speed is not so stable or in remote areas where mobile reception is not so good, but also in cities where building and other construction may block mobile signals. Therefore, designing lightweight networks with low memory requirement and computational cost is one of the most practical solutions for face verification on mobile platform. In this paper, a novel mobile network named SeesawFaceNets, a simple but effective model, is proposed for productively deploying face recognition for mobile devices. Dense experimental results have shown that our proposed model SeesawFaceNets outperforms the baseline MobilefaceNets, with only {\bf66\%}(146M VS 221M MAdds) computational cost, smaller batch size and less training steps, and SeesawFaceNets achieve comparable performance with other SOTA model e.g. mobiface with only {\bf54.2\%}(1.3M VS 2.4M) parameters and {\bf31.6\%}(146M VS 462M MAdds) computational cost, It is also eventually competitive against large-scale deep-networks face recognition on all 5 listed public validation datasets, with {\bf6.5\%}(4.2M VS 65M) parameters and {\bf4.35\%}(526M VS 12G MAdds) computational cost.

Figures

Figures reproduced from arXiv: 1908.09124 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 49 canonical work pages

[1]

S. Chen, Y . Liu, X. Gao, and Z. Han. Mobilefacenets: Efﬁ- cient cnns for accurate real-time face veriﬁcation on mobile devices. arXiv preprint arXiv:1804.07573, 2018

arXiv 2018
[2]

J. Zhang. Seesaw-Net: Convolution Neural Net- work With Uneven Group Convolution. arXiv preprint arXiv:1905.03672, 2019

work page Pith review arXiv 1905
[3]

Ramachandran. P, Zoph. B., Le. Q. V . Searching for acti- vation functions. arXiv preprint arXiv:1710.05941, 2017

arXiv 2017
[4]

J. Deng, J. Guo, N. Xue, S. Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4690-4699, 2019

work page 2019
[5]

MobiFace: A Lightweight Deep Learning Face Recognition on Mobile Devices

C.N. Duong, et al. Mobiface: A lightweight deep learn- ing face recognition on mobile devices. arXiv preprint arXiv:1811.11080, 2018

work page Pith review arXiv 2018
[6]

Searching for mobilenetv3

Howard, A, et al. Searching for mobilenetv3. arXiv preprint arXiv:1905.02244, 2019

arXiv 1905
[7]

J. Hu, L. Shen, G. Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

work page 2018
[8]

Krizhevsky, et al

A. Krizhevsky, et al. One weird trick for parallelizing con- volutional neural networks. arXiv preprint arXiv:1404.5997, 2014

arXiv 2014

Show all 64 references

[9]

Goyal, P

P. Goyal, P. Dollr, R. Girshick, et al. Accurate, large mini- batch sgd: Training imagenet in 1 hour. arXiv preprint arXiv:1706.02677, 2017

2017 arXiv
[10]

Hinton, O

G. Hinton, O. Vinyals, J. Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

2015 arXiv
[11]

Courbariaux and Y

M. Courbariaux and Y . Bengio. Binarynet: Training deep neural networks with weights and activations constrained to +1 or -1. CoRR, abs/1602.02830, 2016

2016 arXiv
[12]

Courbariaux, Y

M. Courbariaux, Y . Bengio, and J. David. Binaryconnect: Training deep neural networks with binary weights during propagations. In NIPS, pages 3123–3131, 2015

2015
[13]

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

2009
[14]

C. N. Duong, K. Luu, K. Quach, and T. Bui. Beyond princi- pal components: Deep boltzmann machines for face model- ing. In CVPR, 2015

2015
[15]

C. N. Duong, K. Luu, K. Quach, and T. Bui. Longitudinal face modeling via temporal deep restricted boltzmann ma- chines. In CVPR, 2016

2016
[16]

C. N. Duong, K. Luu, K. Quach, and T. Bui. Deep appear- ance models: A deep boltzmann machine approach for face modeling. Intl Journal of Computer Vision (IJCV), 2018

2018
[17]

C. N. Duong, K. G. Quach, K. Luu, T. H. N. Le, and M. Sav- vides. Temporal non-volume preserving approach to facial age-progression and age-invariant face recognition. InICCV, 2017

2017
[18]

Girshick, J

R. Girshick, J. Donahue, T. Darrell, and J. Malik. Rich fea- ture hierarchies for accurate object detection and semantic segmentation. 2014

2014
[19]

Y . Guo, L. Zhang, Y . Hu, X. He, and J. Gao. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition. In European Conference on Computer Vision , pages 87–102. Springer, 2016

2016
[20]

Sengupta, J.-C

S. Sengupta, J.-C. Chen, C. Castillo, V . M. Patel, R. Chel- lappa, and D. W. Jacobs. Frontal to proﬁle face veriﬁcation in the wild. In WACV, 2016

2016
[21]

Moschoglou, A

S. Moschoglou, A. Papaioannou, C. Sagonas, J. Deng, I. Kot- sia, and S. Zafeiriou. Agedb: The ﬁrst manually collected in-the-wild age database. In CVPR Workshop, 2017

2017
[22]

Zheng and W

T. Zheng and W. Deng. Cross-pose lfw: A database for studying cross-pose face recognition in unconstrained envi- ronments. Technical Report, 2018

2018
[23]

Zheng, W

T. Zheng, W. Deng, and J. Hu. Cross-age lfw: A database for studying cross-age face recognition in unconstrained en- vironments. arXiv:1708.08197, 2017

2017 arXiv
[24]

M. S. H. N. Le, R. Gummadi. Deep recurrent level set for segmenting brain tumors. In Medical Image Computing and Computer Assisted Intervention (MICCAI) , pages 646–653. Springer, 2018

2018
[25]

S. Han, H. Mao, and W. J. Dally. Deep compression: Com- pressing deep neural network with pruning, trained quanti- zation and huffman coding. CoRR, abs/1510.00149, 2015

2015 arXiv
[26]

S. Han, J. Pool, J. Tran, and W. J. Dally. Learning both weights and connections for efﬁcient neural networks. In Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, NIPS’15, pages 1135–1143, Cambridge, MA, USA, 2015. MIT Press

2015
[27]

K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick. Mask R- CNN. In Proceedings of the International Conference on Computer Vision (ICCV), 2017

2017
[28]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016

2016
[30]

A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. Mobilenets: Efﬁ- cient convolutional neural networks for mobile vision appli- cations. arXiv preprint arXiv:1704.04861, 2017

2017 arXiv
[31]

Y . Wen, K. Zhang, Z. Li and Y . Qiao A discriminative feature learning approach for deep face recognition. InECCV, pages 499–515, 2016

2016
[32]

W. Liu, Y . Wen, Z. Yu, M. Li, B. Raj, and L. Song. Sphereface: Deep hypersphere embedding for face recogni- tion. In CVPR, 2017

2017
[33]

H. Wang, Y . Wang, Z. Zhou, X. Ji, Z. Li, D. Gong, J. Zhou, and W. Liu. Cosface: Large margin cosine loss for deep face recognition. In CVPR, 2018

2018
[34]

C N. Duong. ShrinkTeaNet: Million-scale Lightweight Face Recognition via Shrinking Teacher-Student Networks.arXiv preprint arXiv:1905.10620, 2019

1905 arXiv
[35]

X. Li. AirFace: Lightweight and Efﬁcient Model for Face Recognition. arXiv preprint arXiv:1907.12256, 2019

1907 arXiv
[36]

Zhang, X

X. Zhang, X. Zhou, M. Lin, and J. sun. Shufﬂenet: An ex- tremely efﬁcient convolutional neural network for mobile de- vices. In CVPR, pages 6848–6856, 2018

2018
[37]

Huang, Z

G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger. Densely connected convolutional networks. In CVPR, pages 2261–2269, 2017

2017
[38]

G. B. Huang, M. Mattar, T. Berg, and E. Learned-Miller. La- beled faces in the wild: A database forstudying face recog- nition in unconstrained environments. In Workshop on faces in’Real-Life’Images: detection, alignment, and recognition, 2008

2008
[39]

Hubara, M

I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y . Bengio. Binarized neural networks. InNIPS, pages 4107– 4115, 2016

2016
[40]

H. N. Le, C. N. Duong, K. Luu, and M. Savvides. Deep contextual recurrent residual networks for scene labeling. In Journal of Pattern Recognition, 2018

2018
[41]

H. N. Le, K. G. Quach, K. Luu, and M. Savvides. Reformu- lating level sets as deep recurrent neural network approach to semantic segmentation. In Trans. on Image Processing (TIP), 2018

2018
[42]

H. N. Le, C. Zhu, Y . Zheng, K. Luu, and M. Savvides. Ro- bust hand detection in vehicles. In Intl. Conf. on Pattern Recognition (ICPR), 2016

2016
[43]

Q. Li, S. Jin, and J. Yan. Mimicking very efﬁcient network for object detection. 2017 IEEE Conference on CVPR, pages 7341–7349, 2017

2017
[44]

Z. Liu, J. Li, Z. Shen, G. Huang, S. Yan, and C. Zhang. Learning efﬁcient convolutional networks through network slimming. 2017 IEEE International Conference on Com- puter Vision (ICCV), pages 2755–2763, 2017

2017
[45]

J. Long, E. Shelhamer, and T. Darrell. Fully convolutional networks for semantic segmentation. In CVPR
[46]

Z. Qin, Z. Zhang, X. Chen, C. Wang, and Y . Peng. Fd- mobilenet: Improved mobilenet with a fast downsampling strategy. In 2018 25th IEEE International Conference on Image Processing (ICIP), pages 1363–1367. IEEE, 2018

2018
[47]

Rastegari, V

M. Rastegari, V . Ordonez, J. Redmon, and A. Farhadi. Xnor- net: Imagenet classiﬁcation using binary convolutional neu- ral networks. In ECCV (4), volume 9908 of Lecture Notes in Computer Science, pages 525–542. Springer, 2016

2016
[48]

Sandler, A

M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen. Mobilenetv2: Inverted residuals and linear bottle- necks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4510–4520, 2018

2018
[49]

K. Sun. IGCV3: Interleaved low-rank group convolu- tions for efﬁcient deep neural networks. arXiv preprint arXiv:1806.00178, 2018

2018 arXiv
[50]

M. Tan. Mnasnet: Platform-aware neural architecture search for mobile. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition , pages 2820–2828, 2019

2019
[51]

N. Ma, X. Zhang, H. Zheng, and J. sun. Shufﬂenet v2: Practi- cal guidelines for efﬁcient cnn architecture design. InECCV, pages 116–131, 2018

2018
[52]

Schroff, D

F. Schroff, D. Kalenichenko, and J. Philbin. Facenet: A uniﬁed embedding for face recognition and clustering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 815–823, 2015

2015
[53]

Simonyan and A

K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition, 2014

2014
[54]

H. Wang, Y . Wang, Z. Zhou, X. Ji, D. Gong, J. Zhou, Z. Li, and W. Liu. Cosface: Large margin cosine loss for deep face recognition. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018

2018
[55]

Y . Wei, X. Pan, H. Qin, and J. Yan. Quantization mimic: Towards very tiny cnn for object detection. CoRR, abs/1805.02152, 2018

2018 arXiv
[56]

X. Wu, R. He, Z. Sun, and T. Tan. A light cnn for deep face representation with noisy labels. IEEE Transactions on In- formation Forensics and Security, 13(11):2884–2896, 2018

2018
[57]

Zhang, Z

K. Zhang, Z. Zhang, Z. Li, and Y . Qiao. Joint face detection and alignment using multitask cascaded convolutional net- works. IEEE Signal Processing Letters, 23(10):1499–1503, 2016

2016
[58]

Zheng, C

Y . Zheng, C. Zhu, K. Luu, H. N. Le, C. Bhagavatula, and M. Savvides. Towards a deep learning framework for uncon- strained face detection. In BTAS, 2016

2016
[59]

C. Zhu, Y . Ran, K. Luu, and M. Savvides. Seeing small faces from robust anchor’s perspective. InCVPR, 2018

2018
[60]

C. Zhu, Y . Zheng, K. Luu, H. N. Le, C. Bhagavatula, and M. Savvides. Weakly supervised facial analysis with dense hyper-column features. In IEEE Conference on Computer Vision and Pattern Recognition Workshop (CVPRW), 2016

2016
[61]

C. Zhu, Y . Zheng, K. Luu, and M. Savvides. Enhancing inte- rior and exterior deep facial features for face detection in the wild. In Intl Conf. on Automatic Face and Gesture Recogni- tion (FG), 2018

2018
[62]

F. N. Iandola, M. W. Moskewicz, K. Ashraf et al. . Squeezenet: Alexnet-level accuracy with 50x fewer parame- ters and 1mb model size. arXiv preprint arXiv:1602.07360, 2016

2016 arXiv
[63]

Zhang, J

J. Zhang, J. Zhao, Z. Wang. Analysis of GPU and FPGA parallel computing - example of Kirchhoff prestack time mi- gration. In Progress in Geophysics, 2013

2013
[64]

Ignatov, R

A. Ignatov, R. Timofte, W. Chou, K. Wang, M. Wu, T. Hart- ley, L. Van Gool Ai benchmark: Running deep neural net- works on android smartphones. In ECCV, pages 0–0, 2018

2018
[65]

Paszke, S

A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. De- Vito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer Auto- matic differentiation in pytorch. In NIPS Autodiff Workshop, 2017

2017

Pith tools

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

[1] [1]

S. Chen, Y . Liu, X. Gao, and Z. Han. Mobilefacenets: Efﬁ- cient cnns for accurate real-time face veriﬁcation on mobile devices. arXiv preprint arXiv:1804.07573, 2018

arXiv 2018

[2] [2]

J. Zhang. Seesaw-Net: Convolution Neural Net- work With Uneven Group Convolution. arXiv preprint arXiv:1905.03672, 2019

work page Pith review arXiv 1905

[3] [3]

Ramachandran. P, Zoph. B., Le. Q. V . Searching for acti- vation functions. arXiv preprint arXiv:1710.05941, 2017

arXiv 2017

[4] [4]

J. Deng, J. Guo, N. Xue, S. Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4690-4699, 2019

work page 2019

[5] [5]

MobiFace: A Lightweight Deep Learning Face Recognition on Mobile Devices

C.N. Duong, et al. Mobiface: A lightweight deep learn- ing face recognition on mobile devices. arXiv preprint arXiv:1811.11080, 2018

work page Pith review arXiv 2018

[6] [6]

Searching for mobilenetv3

Howard, A, et al. Searching for mobilenetv3. arXiv preprint arXiv:1905.02244, 2019

arXiv 1905

[7] [7]

J. Hu, L. Shen, G. Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

work page 2018

[8] [8]

Krizhevsky, et al

A. Krizhevsky, et al. One weird trick for parallelizing con- volutional neural networks. arXiv preprint arXiv:1404.5997, 2014

arXiv 2014