Attention Control with Metric Learning Alignment for Image Set-based Recognition

B.V.K Vijaya Kumar; Jane You; Xiaofeng Liu; Zhenhua Guo

REVIEW 3 major objections 5 minor 75 references

Attention Control with Metric Learning Alignment for Image Set-based Recognition

T0 review · 3 major / 5 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 weighting each image in a face set by how it relates to the other images—learned by an actor-critic reinforcement learning agent—outperforms independent per-image quality scoring for set-based and video-based face…

desk verdict The DAC extension is real and the benchmark work is solid, but the paper never specifies a traversal order for orderless sets, and since the MDP state depends on that order, the central permutation-invariance claim is unexamined. read the letter →

arxiv 1908.01872 v1 pith:53P2ZS2X submitted 2019-08-05 cs.CV cs.LGeess.IV

Xiaofeng Liu , Zhenhua Guo , Jane You , B.V.K Vijaya Kumar This is my paper · ORCID

classification cs.CVcs.LGeess.IV

keywords facerecognitionimageset-basedactor-criticreinforcementlearningdependency-awareattentioncontrolpose-guidedrepresentationmetricvideoset-to-set

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 in set-based face recognition, the importance of each image should be decided jointly with the others: a redundant low-quality duplicate should be down-weighted even if it is clear, while a blurry but complementary profile view should be kept. To learn such weights, the authors formulate set aggregation as a Markov decision process (a step-by-step decision problem) in feature space and train a dependency-aware attention control (DAC) network with actor-critic reinforcement learning, using the change in softmax classification loss as reward. They also propose a pose-guided representation that separates frontal and profile images, first with a parameter-free version that needs pose detection and then with a metric-learning version that removes pose detection at test time. Evaluations on IJB-A, IJB-B, IJB-C, YTF, and Celebrity-1000 report higher verification and identification accuracy than independent-quality baselines such as NAN, and the authors describe the results as state of the art. A reader should care because the method removes the need for per-image quality labels and offers a plug-in module that works over any fixed face embedding.

What carries the argument

The load-bearing object is the dependency-aware attention control (DAC) module: a two-branch actor-critic network that acts on 128-dimensional face embeddings. At step $t$, the state is $s_t = \left\{ (\sum_i a_i f_i - f_t)/(\sum_i a_i - 1) \right\}$ concatenated with $f_t$; the actor outputs a continuous attention weight $a_t \in [0,1]$ from a Gaussian policy, and the critic estimates the state value. The reward is $r_t = L_m[h(g(X_m|s_t))] - L_m[h(g(X_m|s_{t+1}))] + \lambda \max[0, (1-a_t)]$: the decrease of the softmax cross-entropy of a classifier on the aggregated feature, plus a hinge penalty that penalizes keeping redundant images. An off-policy variant reuses past experience through importance-sampling ratios and a clipped off-policy correction, and a trust-region constraint keeps the updated policy close to a running average policy to stabilize training. The second mechanism is the pose-guided representation: the parameter-free version splits the set into frontal and profile groups using estimated yaw angles and computes the inter-set distance from group-weighted Euclidean distances, while the metric-learning version learns pose-group centroids and pushes different pose groups apart and same-identity centroids together, so pose detection is not needed at test time.

What would settle it

On one fixed split of IJB-A with a frozen CNN, compare three aggregators on the same features: uniform average, NAN-style independent quality weights, and DAC weights trained with the paper's softmax reward. If DAC does not beat both alternatives on 1:1 verification TAR at FAR=0.01 and open-set rank-1 TPIR at FPIR=0.01 when evaluated with L2 distances, the central claim is unsupported. A complementary check is to retrain DAC with a reward defined by a verification-specific loss such as triplet or contrastive loss on aggregated features; if that reward yields strictly better L2 verification than the softmax reward, the paper's training objective is misaligned with its evaluation metric.

Watch

Extended reading notes

Core claim

The central claim is that an image's weight within a set cannot be assigned in isolation: an image that adds nothing beyond what the set already contains should be down-weighted even if it is high quality, while a complementary but low-quality view should be kept. The paper makes this precise by casting set aggregation as a Markov decision process in latent space and training a dependency-aware attention control (DAC) network with actor-critic reinforcement learning. The agent traverses the set; at each step the state is the current feature concatenated with the weighted aggregation of the remaining features, the action is a continuous weight in [0, 1], and the reward is the reduction of softmax cross-entropy loss caused by the weight update, with a hinge penalty that discourages keeping redundant images. The paper then adds a pose-guided representation that separates frontal and profile images, first using pose detection and later replacing it with learned pose-group centroids and a metric loss, and combines DAC with temporal convolution for video frames. Under this design, the authors report accuracy gains over independent-quality baselines such as NAN on IJB-A/B/C, YTF, and Celebrity-1000, with verification and identification numbers they describe as state of the art.

Load-bearing premise

The load-bearing premise is that improving a softmax classification loss on the aggregated feature also improves the L2-distance-based verification and open-set identification scores that the paper reports, since the reinforcement learning reward is computed from the softmax loss while the evaluation uses L2 distances between aggregated feature vectors.

Editorial extensions

If this is right

Set-level face recognition can be improved without per-image quality labels, since the DAC learns its weights from set-level identity annotations alone.
The method works with sets of variable size because aggregation remains a weighted average inside the convex hull of the feature vectors and can be applied on top of any fixed CNN embedding.
Video-based recognition can use a divide-and-conquer scheme: temporal convolution handles the frame sequences while DAC handles the remaining images, which also lowers the average verification time on IJB-A from 62 ms to 48 ms.
The metric-learning pose alignment removes pose detection at test time while retaining most of the extreme-pose gains, reducing inference cost on IJB-A from 107 ms to 71 ms.
The authors argue the module is a general solution for orderless-sample recognition and could be transferred to person re-identification, action recognition, and event detection.

Reading between the lines

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

Because the reward is based on softmax cross-entropy while the reported open-set results are L2-based, a natural extension is to define the reward directly on a verification loss over aggregated features; the authors note triplet training as possible future work, which would align the training objective with the test metric.
The MDP traverses the set in a fixed order even though the set is described as orderless; a permutation-invariant policy, such as a set transformer or full-set attention, could remove order sensitivity and better match the stated premise.
Since DAC is trained on a frozen embedding, any future upgrade of the feature extractor would change what counts as redundant, so the attention module would likely need retraining after a backbone change.
The pose-guided grouping mechanism only needs a way to partition a set into groups, so the same idea could be applied to other nuisance factors such as illumination or resolution whenever such a partitioning signal is available.

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

The thing to know: this is a legitimate extension of the authors' ECCV 2018 DAC paper, with real new components (ML-PGR, divide-and-conquer temporal combination) and unusually extensive evaluation on IJB-A/B/C, YTF, and Celebrity-1000. But there is one load-bearing gap: the paper calls the input sets "orderless," yet the DAC is trained as an MDP over a sequence, Eq. (1) builds the state from already-updated weights versus still-untouched weights, and no traversal order is ever specified. Permuting the same template changes the state sequence and generally changes the learned weights, so the reported gains over NAN could partly be an artifact of storage order. That needs to be addressed directly, either by randomizing order at train/test and showing invariance, or by fixing and disclosing an order and arguing why the claim survives.

What the paper does well: the empirical comparisons are disciplined. DAC is compared against same-backbone baselines (NAN, AvePool, Mean L2) and shows consistent gains across multiple datasets. The authors honestly report that the softmax-termination variant hurts on IJB-A, which is the kind of negative result that adds credibility. The runtime analysis is genuinely useful, and ML-PGR's removal of pose detection at test time is a practical contribution. The extension over the ECCV preliminary version is clearly itemized rather than buried.

Softer spots, in proportion: the reward/train-test metric mismatch the reader flagged is real but secondary. The reward is softmax cross-entropy while verification/open-set identification uses L2 distances; however, NAN and the other baselines share the same softmax-training/L2-testing split, so the relative comparison can survive it. I would still want an analysis of whether CE improvement transfers to L2 matching. Minor concerns: hyperparameters are tuned per dataset; error bars are missing from several IJB-B/C tables; code availability is promised but not shown. The heavy self-citation pattern is not by itself a flaw here because the prior ECCV DAC is openly acknowledged and the new material is distinct.

Bottom line: the central idea is worth engaging with, but in its current form the permutation question is unresolved. A serious referee should ask for either a permutation-robustness experiment or an explicit ordering protocol, plus code. I would send it to review rather than desk-reject, and I would want the revision to close the order-dependence gap before treating the dependency-aware claim as established.

Referee Report

3 major / 5 minor

Summary. The paper proposes a dependency-aware attention control (DAC) module for image set-based and video-based face verification and identification. DAC casts the sequential assignment of soft attention weights to images in a set as a Markov decision process, solved with an actor-critic reinforcement learning method that includes off-policy experience replay and a trust-region-style KL constraint. The paper also introduces two pose-guided representation (PGR) schemes for inter-set pose dependency, one parameter-free and one based on metric learning, and combines DAC with temporal attention models (RNN or temporal convolution) using a divide-and-conquer strategy. Experiments on IJB-A, IJB-B, IJB-C, YTF, and Celebrity-1000 report consistent gains over same-backbone baselines such as NAN and average pooling.

Significance. If the claims hold, the work addresses a real limitation of prior attention-based set aggregation: independent quality assessment ignores redundancy and inner-set dependency. The experimental design is a strength because the main comparisons (NAN, CNN+AvePool, CNN+Mean L2) use the same backbone, making the reported gains credible. The paper is also honest in reporting that the softmax-based termination condition hurts IJB-A performance. However, the central conceptual claim that the method models dependencies among orderless images is undermined by the absence of any permutation-robustness analysis, since the proposed MDP state and policy are order-dependent. The evidence is extensive but does not fully support the 'orderless' formulation as presented.

major comments (3)

[Section III-A, Eq. (1)] The traversal order for 'orderless' sets is never defined. The state at step t is the concatenation of f_t with (Σ_i a_i f_i − f_t)/(Σ_i a_i − 1), where a_i have already been updated for previously visited images and remain 1 for not-yet-visited ones. Consequently, permuting the same set changes the state sequence and, in general, changes the learned weights and the final aggregated feature. Since templates in IJB-A, IJB-B, IJB-C, YTF, and Celebrity-1000 have no canonical order, the reported improvements over NAN could be partly an artifact of the template storage order rather than of dependency modeling. Please specify how the traversal order is chosen and either enforce permutation invariance or report results over multiple random permutations; if the method remains order-dependent, the 'orderless' claim must be revised.
[Section III-A, Eq. (3) and Section V-A] The DAC reward is defined as the change in softmax cross-entropy loss (plus a hinge penalty), but verification and open-set identification are evaluated with L2 distances between aggregated feature vectors. The paper acknowledges that a softmax-based termination condition degrades IJB-A results, indicating that the learned policy is calibrated to softmax confidence rather than to the L2 matching metric. Because the central claim is improved recognition under the L2 protocol, the paper should analyze whether the CE-based reward transfers to L2-based matching. An ablation using a distance-based reward (e.g., triplet or contrastive loss) on a validation split, or a correlation analysis between the reward and the L2 verification metric, would substantiate the claim.
[Tables I and III, Section V-A] The improvements attributed to the pose-guided representation on the full IJB-A protocol are very small and generally within the reported standard deviations: e.g., rank-1 improves from 0.972±0.012 to 0.973±0.011 for PF-PGR and TPIR@FPIR=0.01 from 0.853±0.033 to 0.855±0.042. The larger gains reported for frontal-only and profile-only probes in Table III have no standard deviations or significance tests. Either add significance testing or restrict the claim of consistent improvement to the frontal/profile-only setting.

minor comments (5)

[Section III-A] The text says the state is 'related to the t−1 weighted features and T−(t−1) to-be-weighted features,' but Eq. (1) actually aggregates all features except f_t with updated weights for visited images and unit weights for unvisited ones. The wording and the equation should be reconciled.
[Section III-C] The divide-and-conquer combination with temporal models does not specify how mixed templates are split into still images and video frames. In IJB-A/B/C, a template can contain both stills and multiple video sequences; the exact partitioning rule affects reproducibility and should be stated.
[Section IV-B, Eq. (12)] The phrase 'with equal weights' is ambiguous: it is not clear how the cross-entropy loss and the metric loss in Eq. (12) are weighted relative to each other, nor how the three terms within the metric loss are combined.
[Section V, Tables IV-VI] The claim of state-of-the-art results is broader than what the comparisons support, since many cited methods use different backbones and pre-training data. The strongest and fairest claim is the improvement over same-backbone baselines such as NAN and the pooling methods.
[Section III-B, Eqs. (9)-(10)] The closed-form solution to the trust-region constrained quadratic program is asserted without derivation. A short derivation or a reference to the precise mechanism would help readers verify the update rule.

Circularity Check

1 steps flagged · score 4.0 of 10

ML-PGR probe-set training makes one extreme-pose result partly self-confirming; core DAC comparisons are independent.

fitted input called prediction [Section V-A, Table III caption/context]
"We note that the ML-PGR is trained with probe set with all images in IJB-A."

The ML-PGR feature extractor is fit with the metric-learning objective (Eq. 12) using the full IJB-A probe set, and Table III then reports open-set identification on only-frontal/only-profile subsets of that same probe set. The reported >2%/>3% TPIR/rank-1 gains of DAC(off)&ML-PGR over DAC(off) are therefore not held-out predictions: the model was optimized on the same probe identities and pose structure it is evaluated on. The improvement is partly by construction, though the parameter-free PF-PGR in the same table provides an uncontaminated version of the extreme-pose boost.

full rationale

The central DAC claim is not circular: the actor-critic policy is trained to maximize a reward defined by change in softmax cross-entropy loss, but the headline verification/identification results on IJB-A/B/C, YTF, and Celebrity-1000 are measured on held-out identities against external baselines (NAN, mean pooling, QAN, etc.), so the main comparisons are not self-defined. The reward/L2-testing mismatch is a real train/test metric gap, but the paper explicitly acknowledges that softmax-based reward is only used in closed-set testing and cannot be applied to open-set identification; this is an external-validity concern, not circularity. The unspecified traversal order for orderless sets and the resulting order-dependence of Eq. (1) is a robustness/invariance flaw, but it does not make the prediction equivalent to its inputs by construction, so it is not scored as circularity. The one genuine circular step is confined to the ML-PGR extreme-pose ablation: the metric-learning feature extractor is explicitly trained on the IJB-A probe set before being evaluated on subsets of that same probe set. Because the parameter-free PF-PGR gives similar gains without any training on the probe set, the broad PGR idea retains independent support; the self-confirming component is the ML-PGR variant's reported improvement in Table III. Accordingly, the paper receives a partial circularity score of 4 rather than a higher score reserved for central claims that reduce to a fit or self-citation chain.

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

No new physical or formal entities are postulated; the proposed DAC and PGR are trainable modules with direct empirical evaluation on benchmarks. The central claim rests on assumptions about the reward signal, the permutation invariance of the sequential attention process, and the sufficiency of coarse pose grouping. These are standard engineering choices, but they are not proven and not all are tested by ablation.

free parameters (5)

lambda = 0.1 (IJB-A/B/C), 0.12 (YTF, Celebrity-1000)
Hinge penalty weight in the DAC reward (Eq. 3), chosen empirically per dataset; directly controls the redundancy reduction trade-off.
beta = 1
Threshold for pose group separation in the ML-PGR loss (Eq. 12); set by hand without sensitivity analysis.
phi = 5
Threshold for probe-gallery distance in the ML-PGR loss (Eq. 12); set by hand without sensitivity analysis.
termination_threshold = 0.6 (IJB-A), 0.5 (Celebrity-1000)
Softmax prediction threshold for early termination in closed-set identification; 'validated' on each dataset, and its use hurts IJB-A training.
KL_bound_xi = not reported
KL divergence bound xi in Eq. (9) for the trust-region update; the value is never disclosed.

assumptions (5)

domain assumption The softmax cross-entropy loss difference used as reward (Eq. 3) is a valid objective for improving L2-distance-based verification and open-set identification metrics.
Training uses CE reward; evaluation uses L2 distances on aggregated features; the paper provides no formal or empirical analysis of this transfer.
domain assumption The sequential MDP traversal is invariant to the order of images in the set, so the final attention weights are meaningful for orderless sets.
Image sets are orderless by definition, but DAC processes images in a sequence with a policy that depends on the current state; the paper does not specify the traversal order or test permutation invariance.
domain assumption Coarse pose grouping (frontal at most 30 degrees, profile greater than 30 degrees, left/right mirrored) is sufficient to model inter-set pose dependency.
PF-PGR and ML-PGR discretize continuous pose into two or three groups; no evidence that this granularity preserves the information needed for extreme-pose matching.
standard math The off-policy trust-region update (Eqs. 9-10) preserves convergence and stability of the actor-critic.
The update is adapted from published TRPO/ACER derivations; the paper cites [58], [64] and does not prove convergence, treating it as standard RL machinery.
domain assumption The fixed pretrained CNN embeddings contain enough identity information for the attention module to make beneficial weighting decisions.
The DAC is trained on frozen features from GoogLeNet/ResNet; the reported gains depend on the quality and completeness of these features, which are not fine-tuned except in ML-PGR.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Attention Control with Metric Learning Alignment for Image Set-based Recognition." pith.science (2026). https://pith.science/paper/53P2ZS2X

@misc{pith2026190801872,
  author       = {Pith},
  title        = {Pith review of: Attention Control with Metric Learning Alignment for Image Set-based Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/53P2ZS2X}},
  note         = {Machine review of arXiv:1908.01872}
}

abstract

This paper considers the problem of image set-based face verification and identification. Unlike traditional single sample (an image or a video) setting, this situation assumes the availability of a set of heterogeneous collection of orderless images and videos. The samples can be taken at different check points, different identity documents $etc$. The importance of each image is usually considered either equal or based on a quality assessment of that image independent of other images and/or videos in that image set. How to model the relationship of orderless images within a set remains a challenge. We address this problem by formulating it as a Markov Decision Process (MDP) in a latent space. Specifically, we first propose a dependency-aware attention control (DAC) network, which uses actor-critic reinforcement learning for attention decision of each image to exploit the correlations among the unordered images. An off-policy experience replay is introduced to speed up the learning process. Moreover, the DAC is combined with a temporal model for videos using divide and conquer strategies. We also introduce a pose-guided representation (PGR) scheme that can further boost the performance at extreme poses. We propose a parameter-free PGR without the need for training as well as a novel metric learning-based PGR for pose alignment without the need for pose detection in testing stage. Extensive evaluations on IJB-A/B/C, YTF, Celebrity-1000 datasets demonstrate that our method outperforms many state-of-art approaches on the set-based as well as video-based face recognition databases.

Figures

Figures reproduced from arXiv: 1908.01872 by the authors.

**Figure 2.** Typical examples in the test set of (a) YTF and (b) IJB-A dataset showing the weights ( [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Our network architecture for image set-based face recognition. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 5.** Figure 5: Illustration of the temporal attention scheme which [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: Illustration of the parameter-free pose-guided represen [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

**Figure 7.** Figure 7: Illustration of the metric learning based pose-guided representation scheme. Left: training stage. Right: testing stage. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

**Figure 8.** Figure 8: Average ROC (Left) and CMC (Right) curves of the [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

**Figure 9.** Figure 9: The CMC curves of different methods on Celebrity [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

**Figure 8.** Figure 8: Performance comparation on (a) IJB-A, (b) YTF, and (c) Celebrity-1000 datasets. The larger episode rewards, the bette Fig. 10: Performance comparison on (a) IJB-A, (b) YTF, and (c) Celebrity-1000 datasets. The larger episode rewards, the better [PITH_FULL_IMAGE:figure…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

75 extracted references · 65 canonical work pages

[1]

S.-I. Amari. Natural gradient works efﬁciently in learning. Neural computation, 10(2):251–276, 1998

work page 1998
[2]

A. M. Andrew. Reinforcement learning: An introduction by richard s. sutton and andrew g. barto, adaptive computation and machine learning series, mit press (bradford book), cambridge, mass., 1998, xviii+ 322 pp, isbn 0-262-19398-1,(hardback,£ 31.95).-. Robotica, 17(2):229–235, 1999

work page 1998
[3]

Q. Cao, L. Shen, W. Xie, O. M. Parkhi, and A. Zisserman. Vggface2: A dataset for recognising faces across pose and age. In Automatic Face & Gesture Recognition (FG 2018), 2018 13th IEEE International Conference on, pages 67–74. IEEE, 2018

work page 2018
[4]

Cevikalp and B

H. Cevikalp and B. Triggs. Face recognition based on image sets. In Computer Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on, pages 2567–2573. IEEE, 2010

work page 2010
[5]

D. Chen, S. Ren, Y . Wei, X. Cao, and J. Sun. Joint cascade face detection and alignment. In European Conference on Computer Vision , pages 109–122. Springer, 2014

work page 2014
[6]

J.-C. Chen, R. Ranjan, A. Kumar, C.-H. Chen, V . M. Patel, and R. Chellappa. An end-to-end system for unconstrained face veriﬁcation with deep convolutional neural networks. In IEEE CVPRW, pages 118– 126, 2015

work page 2015
[7]

Cheng, J

G. Cheng, J. Han, P. Zhou, and D. Xu. Learning rotation-invariant and ﬁsher discriminative convolutional neural networks for object detection. IEEE Transactions on Image Processing , 28(1):265–278, 2018. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 13 Method 1:1 Veriﬁcation TAR FAR=1E-5 FAR=1E-4 FAR=1E-3 FAR=1E-2 FAR=1E-1 Caoetal. [3] 0.647 0.7...

work page 2018
[8]

Cheng, C

G. Cheng, C. Yang, X. Yao, L. Guo, and J. Han. When deep learning meets metric learning: Remote sensing image scene classiﬁcation via learning discriminative cnns. IEEE transactions on geoscience and remote sensing, 56(5):2811–2821, 2018

work page 2018

Show all 75 references

[9]

Cheng, P

G. Cheng, P. Zhou, and J. Han. Duplex metric learning for image set classiﬁcation. IEEE Transactions on Image Processing , 27(1):281–292, 2017

2017
[10]

A. R. Chowdhury, T.-Y . Lin, S. Maji, and E. Learned-Miller. One-to- many face recognition with bilinear cnns. In WACV, pages 1–9. IEEE, 2016

2016
[11]

Crosswhite, J

N. Crosswhite, J. Byrne, C. Stauffer, O. Parkhi, Q. Cao, and A. Zis- serman. Template adaptation for face veriﬁcation and identiﬁcation. In FG, pages 1–8. IEEE, 2017

2017
[12]

Ding and D

C. Ding and D. Tao. Trunk-branch ensemble convolutional neural networks for video-based face recognition. In IEEE transactions on pattern analysis and machine intelligence , 2017

2017
[13]

Gao and R

J. Gao and R. Nevatia. Revisiting temporal modeling for video-based person reid. arXiv preprint arXiv:1805.02104 , 2018

2018 arXiv
[14]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 770–778, 2016

2016
[15]

Huang, S

C. Huang, S. Lucey, and D. Ramanan. Learning policies for adaptive tracking with deep feature cascades. arXiv preprint arXiv:1708.02973 , 2017

2017 arXiv
[16]

G. B. Huang, M. Ramesh, T. Berg, and E. Learned-Miller. Labeled faces in the wild: A database for studying face recognition in unconstrained environments. Technical report, Technical Report 07-49, University of Massachusetts, Amherst, 2007

2007
[17]

Ioffe and C

S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning , pages 448–456, 2015

2015
[18]

Janisch, T

J. Janisch, T. Pevn `y, and V . Lis `y. Classiﬁcation with costly features using deep reinforcement learning. arXiv preprint arXiv:1711.07364 , 2017

2017 arXiv
[19]

Jourabloo and X

A. Jourabloo and X. Liu. Pose-invariant face alignment via cnn-based dense 3d model ﬁtting. International Journal of Computer Vision , 124(2):187–203, 2017

2017
[20]

B. F. Klare, B. Klein, E. Taborsky, A. Blanton, J. Cheney, K. Allen, P. Grother, A. Mah, and A. K. Jain. Pushing the frontiers of uncon- strained face detection and recognition: Iarpa janus benchmark a. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recog...

1931
[21]

X. Lan, H. Wang, S. Gong, and X. Zhu. Identity alignment by noisy pixel removal. arXiv preprint arXiv:1707.02785 , 2017

2017 arXiv
[22]

H. Li, G. Hua, X. Shen, Z. Lin, and J. Brandt. Eigen-pep for video face recognition. In Asian Conference on Computer Vision , pages 17–33. Springer, 2014

2014
[23]

Y . Li. Deep reinforcement learning: An overview. arXiv preprint arXiv:1701.07274, 2017

2017 arXiv
[24]

Y . Li, B. Zhang, S. Shan, X. Chen, and W. Gao. Bagging based efﬁcient kernel ﬁsher discriminant analysis for face recognition. In Pattern Recognition, 2006. ICPR 2006. 18th International Conference on, volume 3, pages 523–526. IEEE, 2006

2006
[25]

T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y . Tassa, D. Silver, and D. Wierstra. Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971 , 2015

2015 arXiv
[26]

L.-J. Lin. Self-improving reactive agents based on reinforcement learning, planning and teaching. Machine learning , 8(3-4):293–321, 1992

1992
[27]

L. Liu, L. Zhang, H. Liu, and S. Yan. Toward large-population face identiﬁcation in unconstrained videos. IEEE Transactions on Circuits and Systems for Video Technology , 24(11):1874–1884, 2014

2014
[28]

W. Liu, Y . Wen, Z. Yu, M. Li, B. Raj, and L. Song. Sphereface: Deep hypersphere embedding for face recognition. In IEEE CVPR, volume 1, 2017

2017
[29]

X. Liu. Research on the technology of deep learning based face image recognition. In Thesis, 2019

2019
[30]

X. Liu, Y . Ge, C. Yang, and P. Jia. Adaptive metric learning with deep neural networks for video-based facial expression recognition. Journal of Electronic Imaging , 27(1):013022, 2018

2018
[31]

X. Liu, Z. Guo, S. Li, P. Jia, J. You, and K. B.V .K. Permutation-invariant feature restructuring for correlation-aware image set-based recognition. ICCV 2019

2019
[32]

X. Liu, L. Kong, Z. Diao, and P. Jia. Line-scan system for continuous hand authentication. Optical Engineering , 56(3):033106, 2017

2017
[33]

X. Liu, B. V . Kumar, Y . Ge, C. Yang, J. You, and P. Jia. Normalized face image generation with perceptron generative adversarial networks. In 2018 IEEE 4th International Conference on Identity, Security, and Behavior Analysis (ISBA) , pages 1–8. IEEE, 2018

2018
[34]

X. Liu, B. V . Kumar, P. Jia, and J. You. Hard negative generation for identity-disentangled facial expression recognition. Pattern Recognition, 88:1–12, 2019

2019
[35]

X. Liu, B. V . Kumar, C. Yang, Q. Tang, and J. You. Dependency-aware attention control for unconstrained face recognition with image sets. In European Conference on Computer Vision , pages 573–590. Springer, 2018

2018
[36]

X. Liu, B. V . Kumar, J. You, and P. Jia. Adaptive deep metric learning for identity-aware facial expression recognition. In 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , pages 522–531. IEEE, 2017

2017
[37]

X. Liu, S. Li, L. Kong, W. Xie, P. Jia, J. You, and B. Kumar. Feature- level frankenstein: Eliminating variations for discriminative recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 637–646, 2019

2019
[38]

X. Liu, Z. Li, L. Kong, Z. Diao, J. Yan, Y . Zou, C. Yang, P. Jia, and J. You. A joint optimization framework of low-dimensional projection and collaborative representation for discriminative classiﬁcation. In 2018 24th International Conference on Pattern Recognition (ICPR) , ...

2018
[39]

X. Liu, C. Yang, J. You, J. K. C.-C., and B. Vijaya Kumar. Mutual information regularized feature-level frankenstein for discriminative recognition. In ArXiv, 2019

2019
[40]

X. Liu, Y . Zou, T. Che, P. Jia, J. You, and K. B.V .K. Conservative wasserstein training for pose estimation. In ICCV. IEEE, 2019

2019
[41]

X. Liu, Y . Zou, L. Kong, Z. Diao, J. Yan, J. Wang, S. Li, P. Jia, and J. You. Data augmentation via latent space interpolation for image classiﬁcation. In 2018 24th International Conference on Pattern Recognition (ICPR) , pages 728–733. IEEE, 2018

2018
[42]

X. Liu, Y . Zou, Y . Song, C. Yang, J. You, and B. K Vijaya Kumar. Ordinal regression with neuron stick-breaking for medical diagnosis. In Proceedings of the European Conference on Computer Vision (ECCV) , pages 0–0, 2018

2018
[43]

Y . Liu, J. Yan, and W. Ouyang. Quality aware network for set to set recognition. In Proc. IEEE Int. Conf. Comput. Vision Pattern Recognit. , pages 5790–5799, 2017

2017
[44]

I. Masi, S. Rawls, G. Medioni, and P. Natarajan. Pose-aware face recognition in the wild. In IEEE CVPR , pages 4838–4846, 2016

2016
[45]

I. Masi, A. T. Tráž ˘gn, T. Hassner, J. T. Leksut, and G. Medioni. Do we really need to collect millions of faces for effective face recognition? In ECCV, pages 579–596. Springer, 2016

2016
[46]

B. Maze, J. Adams, J. A. Duncan, N. Kalka, T. Miller, C. Otto, A. K. Jain, W. T. Niggel, J. Anderson, J. Cheney, et al. Iarpa janus benchmark– c: Face dataset and protocol. In 11th IAPR International Conference on Biometrics, 2018

2018
[47]

Meuleau, L

N. Meuleau, L. Peshkin, L. P. Kaelbling, and K.-E. Kim. Off-policy policy search. In MIT Articical Intelligence Laboratory , 2000

2000
[48]

V . Mnih, A. P. Badia, M. Mirza, A. Graves, T. Lillicrap, T. Harley, D. Silver, and K. Kavukcuoglu. Asynchronous methods for deep rein- IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 14 forcement learning. In International Conference on Machine Learning , pages 1928–1...

1928
[49]

V . Mnih, N. Heess, A. Graves, et al. Recurrent models of visual attention. In Advances in neural information processing systems , pages 2204– 2212, 2014

2014
[50]

V . Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529, 2015

2015
[51]

O. M. Parkhi, A. Vedaldi, A. Zisserman, et al. Deep face recognition. In BMVC, volume 1, page 6, 2015

2015
[52]

Peters and S

J. Peters and S. Schaal. Policy gradient methods for robotics. In Intelli- gent Robots and Systems, 2006 IEEE/RSJ International Conference on , pages 2219–2225. IEEE, 2006

2006
[53]

Precup, R

D. Precup, R. S. Sutton, and S. Dasgupta. Off-policy temporal-difference learning with function approximation. In ICML, pages 417–424, 2001

2001
[54]

Y . Rao, J. Lu, and J. Zhou. Attention-aware deep reinforcement learning for video face recognition. In IEEE ICCV , pages 3931–3940, 2017

2017
[55]

Y . Rao, J. Lu, and J. Zhou. Learning discriminative aggregation network for video-based face recognition and person re-identiﬁcation. International Journal of Computer Vision , pages 1–18, 2018

2018
[56]

S. Ren, X. Cao, Y . Wei, and J. Sun. Face alignment at 3000 fps via regressing local binary features. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 1685–1692, 2014

2014
[57]

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
[58]

Schulman, S

J. Schulman, S. Levine, P. Abbeel, M. Jordan, and P. Moritz. Trust region policy optimization. In ICML, pages 1889–1897, 2015

2015
[59]

R. S. Sutton, D. A. McAllester, S. P. Singh, and Y . Mansour. Policy gra- dient methods for reinforcement learning with function approximation. In NIPS, pages 1057–1063, 2000

2000
[60]

Szegedy, V

C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 2818–2826, 2016

2016
[61]

Taigman, M

Y . Taigman, M. Yang, M. Ranzato, and L. Wolf. Deepface: Closing the gap to human-level performance in face veriﬁcation. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 1701–1708, 2014

2014
[62]

D. Tran, L. Bourdev, R. Fergus, L. Torresani, and M. Paluri. Learning spatiotemporal features with 3d convolutional networks. In Proceedings of the IEEE international conference on computer vision , pages 4489– 4497, 2015

2015
[63]

D. Wang, C. Otto, and A. K. Jain. Face search at scale: 80 million gallery. arXiv preprint arXiv:1507.07242 , 2015

2015 arXiv
[64]

Z. Wang, V . Bapst, N. Heess, V . Mnih, R. Munos, K. Kavukcuoglu, and N. de Freitas. Sample efﬁcient actor-critic with experience replay. ICLR, 2017

2017
[65]

Y . Wen, K. Zhang, Z. Li, and Y . Qiao. A discriminative feature learning approach for deep face recognition. In European Conference on Computer Vision , pages 499–515. Springer, 2016

2016
[66]

Whitelam, E

C. Whitelam, E. Taborsky, A. Blanton, B. Maze, J. C. Adams, T. Miller, N. D. Kalka, A. K. Jain, J. A. Duncan, and K. Allen. Iarpa janus benchmark-b face dataset. In CVPR Workshops, 2017

2017
[67]

R. J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. In Reinforcement Learning, pages 5–32. Springer, 1992

1992
[68]

L. Wolf, T. Hassner, and I. Maoz. Face recognition in unconstrained videos with matched background similarity. In Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on , pages 529– 534, 2011

2011
[69]

W. Xie, L. Shen, and A. Zisserman. Comparator networks. In Proceedings of the European Conference on Computer Vision (ECCV) , pages 782–797, 2018

2018
[70]

Xie and A

W. Xie and A. Zisserman. Multicolumn networks for face recognition. arXiv preprint arXiv:1807.09192 , 2018

2018 arXiv
[71]

J. Yang, P. Ren, D. Zhang, D. Chen, F. Wen, H. Li, and G. Hua. Neural aggregation network for video face recognition. In IEEE CVPR , pages 4362–4371, 2017

2017
[72]

Yin and X

X. Yin and X. Liu. Multi-task convolutional neural network for pose- invariant face recognition. In IEEE Transactions on Image Processing , volume 27, pages 964–975. IEEE, 2018

2018
[73]

Zhang, N

J. Zhang, N. Wang, and L. Zhang. Multi-shot pedestrian re-identiﬁcation via sequential decision making. arXiv preprint arXiv:1712.07257 , 2017

2017 arXiv
[74]

Zhang, M

Y . Zhang, M. Pezeshki, P. Brakel, S. Zhang, C. L. Y . Bengio, and A. Courville. Towards end-to-end speech recognition with deep con- volutional neural networks. arXiv preprint arXiv:1701.02720 , 2017

2017 arXiv
[75]

Z. Zhou, Y . Huang, W. Wang, L. Wang, and T. Tan. See the forest for the trees: Joint spatial and temporal recurrent neural networks for video-based person re-identiﬁcation. In Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on , pages 6776–6785. IEEE, 2017

2017

Pith tools

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

[1] [1]

S.-I. Amari. Natural gradient works efﬁciently in learning. Neural computation, 10(2):251–276, 1998

work page 1998

[2] [2]

A. M. Andrew. Reinforcement learning: An introduction by richard s. sutton and andrew g. barto, adaptive computation and machine learning series, mit press (bradford book), cambridge, mass., 1998, xviii+ 322 pp, isbn 0-262-19398-1,(hardback,£ 31.95).-. Robotica, 17(2):229–235, 1999

work page 1998

[3] [3]

Q. Cao, L. Shen, W. Xie, O. M. Parkhi, and A. Zisserman. Vggface2: A dataset for recognising faces across pose and age. In Automatic Face & Gesture Recognition (FG 2018), 2018 13th IEEE International Conference on, pages 67–74. IEEE, 2018

work page 2018

[4] [4]

Cevikalp and B

H. Cevikalp and B. Triggs. Face recognition based on image sets. In Computer Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on, pages 2567–2573. IEEE, 2010

work page 2010

[5] [5]

D. Chen, S. Ren, Y . Wei, X. Cao, and J. Sun. Joint cascade face detection and alignment. In European Conference on Computer Vision , pages 109–122. Springer, 2014

work page 2014

[6] [6]

J.-C. Chen, R. Ranjan, A. Kumar, C.-H. Chen, V . M. Patel, and R. Chellappa. An end-to-end system for unconstrained face veriﬁcation with deep convolutional neural networks. In IEEE CVPRW, pages 118– 126, 2015

work page 2015

[7] [7]

Cheng, J

G. Cheng, J. Han, P. Zhou, and D. Xu. Learning rotation-invariant and ﬁsher discriminative convolutional neural networks for object detection. IEEE Transactions on Image Processing , 28(1):265–278, 2018. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 13 Method 1:1 Veriﬁcation TAR FAR=1E-5 FAR=1E-4 FAR=1E-3 FAR=1E-2 FAR=1E-1 Caoetal. [3] 0.647 0.7...

work page 2018

[8] [8]

Cheng, C

G. Cheng, C. Yang, X. Yao, L. Guo, and J. Han. When deep learning meets metric learning: Remote sensing image scene classiﬁcation via learning discriminative cnns. IEEE transactions on geoscience and remote sensing, 56(5):2811–2821, 2018

work page 2018