Globally Correlation-Aware Hard Negative Generation

Gang Dai; Hongxiang Huang; Quhui Ke; Shuangping Huang; Tianshui Chen; Wenjie Peng

REVIEW 4 major objections 5 minor 84 references

Globally Correlation-Aware Hard Negative Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper argues that synthetic hard negatives for deep metric learning should be generated from correlations measured across the whole mini-batch, not just between a few selected samples.

desk verdict Solid empirical HNG paper with thorough experiments, but the 'global' claim really means global within a single mini-batch, and the promised proof in Section 4.6 is an empirical analysis, not a proof. read the letter →

arxiv 2411.13145 v1 pith:LEO2GLXJ submitted 2024-11-20 cs.CV

Wenjie Peng , Hongxiang Huang , Tianshui Chen , Quhui Ke , Gang Dai , Shuangping Huang This is my paper · ORCID

classification cs.CV

keywords hardnegativegenerationdeepmetriclearninggraphneuralnetworkglobalsamplecorrelationimageretrievalchannel-adaptiveinterpolationhardness-awareloss

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

GCA-HNG proposes that hard negative generation for deep metric learning should be guided by correlations measured across the whole mini-batch rather than by local pair or triplet relationships. The method builds a structured graph whose nodes are sample embeddings and whose edges are pairwise correlations, then runs iterative node and edge message propagation so every edge accumulates global context. Those refined edge representations become per-channel interpolation coefficients that blend an anchor with several negatives from a class, producing negatives whose hardness adapts to how close the negative class is to the anchor class in the batch geometry. The authors report that the resulting synthetic negatives are harder and more diverse, and that the framework improves retrieval on CUB-200-2011, Cars196, Stanford Online Products, and InShop across several backbones and metric losses.

What carries the argument

The load-bearing object is the structured graph $G = (V, E)$ together with its alternating message-passing updates: node message propagation (masked multi-head self-attention plus edge-to-node aggregation, equations 1-2) and edge message propagation (cross-attention from endpoint nodes to the edge, equations 3-4). The refined edge representation $E^K_{ij}$ is turned into a channel-wise interpolation vector via $\lambda_{ij} = \mathrm{Sigmoid}(\mathrm{FC}(E^K_{ij}))$, and the CACAI module uses it in the interpolation formula with the dynamic interval $[0, \eta(d^- - d^+)/d^-]$ controlled by the average metric loss; random weighting across interpolated negatives of the same class provides diversity.

What would settle it

Train on a large-class dataset such as Stanford Online Products with a sampler that deliberately prevents the most confusable class pairs from appearing in the same mini-batch; if Recall@1 stays as high as under random sampling, the batch-level global correlations cannot be the mechanism driving the reported gains, whereas a drop would confirm the paper's own stated limitation.

Watch

Extended reading notes

Core claim

The central discovery is that a global view of sample correlations, obtained by propagating messages through a mini-batch graph, is sufficient to calibrate the hardness of synthetic negatives more accurately than local pair or triplet fusion. In the GCL module, masked multi-head self-attention lets each anchor attend only to negative samples, while edge-to-node aggregation feeds pairwise information into nodes; edge message propagation then uses cross-attention so each edge absorbs the updated context of both endpoints. After K such iterations, the edge representation is mapped through a fully connected layer and sigmoid to a channel-adaptive interpolation vector, and the synthetic negative is computed by interpolating the anchor toward the negative with a dynamic scaling factor that tightens the allowed hardness as average loss decreases. The paper argues that because the synthetic representation is used directly in metric loss optimization, no additional generator network is needed, which makes the method work across CNNs with average pooling, max pooling, and ViT backbones.

Load-bearing premise

The framework assumes that the correlations visible inside one mini-batch of roughly 27 or 40 classes suffice to gauge how hard a synthetic negative should be, so classes that are never sampled together cannot receive the hard negatives they need.

Editorial extensions

If this is right

Hard negative generation can be decoupled from generator networks, since learned interpolation coefficients can feed the metric loss directly.
Hardness becomes batch-geometry-adaptive: closely related classes receive harder synthetic negatives, while distant classes are interpolated more conservatively so synthetic samples stay inside their class region.
The framework plugs into both N-pair and Proxy Anchor losses and multiple backbone families, making it a training-time component rather than a new architecture.
On the four benchmarks, the reported Recall@1, R-precision, and M@R gains come at no inference-time cost, since the graph module is used only during training.
If the global-correlation account is right, the quality of synthetic negatives should improve as the number of classes per batch grows, up to the hardware limit.

Reading between the lines

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

We would expect the channel-adaptive interpolation to also act as a regularizer on embedding geometry, so gains might persist even if synthetic negatives are not explicitly consumed by the loss; a test would be to use the interpolation vectors purely as an auxiliary training signal.
The batch-scope limitation suggests a natural extension: replace the mini-batch graph with a memory-bank graph that maintains correlations across many recent batches, which should matter most on SOP and other large-class datasets.
The same edge-based interpolation coefficients could be interpreted as a learned embedding-space augmentation policy, connecting this work to mixup-style methods, with the difference that mixing coefficients are conditioned on global context rather than sampled randomly.
Because the framework relies on balanced sampling with many classes per batch, its benefit may shrink on long-tailed or extremely high-class datasets, where the hardest distinctions are precisely the ones that rarely co-occur.

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

Short version: this is a solid empirical metric-learning paper, and the gains are probably real, but the title oversells the "global" part. The graph is global over one mini-batch, not over the dataset, and the authors admit in Section 5 that the classes which most need separating often never appear in the same batch. The proof promised in Section 3.3 also never materializes; Section 4.6 gives an empirical analysis instead. Neither flaw kills the paper, but they cap how much weight you should put on the headline claim.

What is genuinely new: the GCL module's iterative node and edge message passing, combined with channel-adaptive interpolation, is a real departure from the local pair or triplet fusion in DAMML, HDML, and Symm. The framework also drops the separate generator network, and Table 8 shows this pays off: a generator-based baseline degrades on max-pooling and ViT features, while GCA-HNG stays consistently strong. That table is the most convincing evidence in the paper. The ablations are also thorough—each module, each loss component, hyperparameter sweeps, sampling strategy comparisons, and structure exploration, all with standard deviations. Code and trained models are promised, which is good practice.

Where the soft spots actually are, in proportion. First, the central claim is over-scoped. The graph is built from one balanced mini-batch of 27 or 40 classes. On SOP, with 11,318 training classes, most class pairs are essentially never co-sampled, so for those pairs the GCL module has no signal and cannot synthesize the hardest negatives. The authors' concession in Section 5 is honest, but it means the SOP and InShop gains should be read as intra-batch improvements, not as confirmation of a truly global correlation mechanism. Second, the promised proof is missing; Section 4.6 is a reasonable empirical argument, not a proof, and the text should say so. Third, using the test set for hyperparameter selection is transparently acknowledged but still a weakness; a validation split or at least a sensitivity analysis on the chosen hyperparameters would strengthen the claims.

Who this is for: anyone working on deep metric learning, hard negative mining, or embedding-space augmentation. It deserves a serious referee. The referee should push for a validation protocol, a real proof or a rewording of the promise, and a discussion of the batch-scope limitation. I would recommend sending it to peer review rather than desk rejection.

Referee Report

4 major / 5 minor

Summary. The paper proposes GCA-HNG, a hard negative generation framework for deep metric learning. It constructs a fully connected graph over a balanced mini-batch, iteratively propagates node and edge messages using masked multi-head self-attention and cross-attention, and produces synthetic negatives by channel-adaptive interpolation between an anchor and multiple negatives from a class. Training proceeds in two stages: first a graph network is optimized with classification, similarity, and diversity losses; then synthetic negatives are integrated into N-pair or Proxy-Anchor metric losses. The method is evaluated on CUB-200-2011, Cars196, SOP, and InShop with GoogLeNet, ResNet-50, and DINO backbones, and the paper reports consistent improvements over baselines, extensive ablations, qualitative visualizations, and released code.

Significance. If the empirical claims hold, the paper makes a useful contribution to hard negative generation: it removes the need for a separate generator network, works across pooling variants and ViT features, and provides a detailed ablation study with standard deviations. The reproducibility package (code and trained models) and the breadth of backbone/loss combinations are strengths. However, the central 'global correlation' claim is only realized within a single mini-batch, and the paper itself concedes this limitation; additionally, the promised proof in Section 4.6 is missing and one key equation is described inconsistently with the text. The contribution is incremental rather than foundational, but it is a reasonable journal-level empirical study if the scope claims are corrected.

major comments (4)

[5 and 3.1] The central claim of a 'global, comprehensive perspective' is not supported by the actual construction. The correlation graph is built from one balanced mini-batch of N times m samples, with N=27 for CUB/Cars and N=40 for SOP/InShop. On SOP, with 11,318 training classes and 40-class batches, the probability that a specific pair of classes is co-sampled in one iteration is about (40/11318)^2, which is around 1.2e-5; over 75k iterations the expected number of co-occurrences for an average class pair is below one, so for most class pairs the GCL module has no meaningful correlation signal, and GCA-HNG cannot synthesize a hard negative between those categories. This is precisely the limitation conceded in Section 5. The title and abstract should be reframed to 'intra-batch' or 'batch-global' correlations, or the method should be extended, for example with a memory bank of class-level proxies, to access cross-batch class relationships. A concrete test would be to compose validation batches from the most confusable class pairs and measure whether the observed gains concentrate there; without such evidence, the improvements on SOP and InShop may reflect only intra-batch negative interactions.
[3.3 and 4.6] Section 3.3 promises 'a detailed proof' in Section 4.6 that directly using interpolated representations avoids the generator-optimization difficulties of prior HNG methods. Section 4.6 contains only an empirical feature-variance analysis and a qualitative argument; no formal proof is given. Either provide an actual proof or remove the promise, since the claim is currently unsupported.
[3.3, Eq. (6)] The interpolation interval stated in the text, [0, eta(d- - d+)/d-], does not match Eq. (6). With d- > d+, Eq. (6) places the synthetic point at fraction (d+ + lambda*eta*(d- - d+))/d- along the anchor-negative direction, so the attainable range is [d+/d-, (d+ + eta*(d- - d+))/d-], not [0, eta*(d- - d+)/d-]. Consequently the synthetic negative is never closer to the anchor than the positive distance d+. Please correct the interval description and state the intended hardness range explicitly in terms of distance from the anchor, as this directly affects the hardness-calibration behavior of the method.
[4.1.1] The test set is used for hyperparameter selection, as disclosed in Section 4.1.1. This is a well-known threat to the validity of comparative claims, and because the baseline numbers from prior work may not have been tuned under the same protocol, the consistent improvements in Tables 1 and 2 could partly reflect test-set tuning. Please report at least one configuration evaluated on a properly held-out validation split (for example, following Musgrave et al.), or provide a sensitivity analysis demonstrating that the main conclusions are stable across a range of hyperparameter choices.

minor comments (5)

[1] In the first paragraph of the introduction, 'boarder range of negative classes' should be 'broader range of negative classes'.
[Figure 2 caption] The caption contains 'Globally ??????????? Learning', indicating a missing or corrupted label in the figure. Please fix the rendering.
[Algorithm 1] The notation sg(.) is used in Algorithm 1 but never defined in the main text or caption; please define it explicitly as stop-gradient.
[3.4, Eq. (11)] The summation from n=l1 to lN is confusing because n denotes a class label, not a running index. Please write the sum over the set of N classes in the batch, for example, sum over n in the sampled class set with n different from l_i, and define that set explicitly.
[3.4, Eq. (14)] Equation (14) uses juxtaposition such as z_i z_hat_in for what appear to be inner products. Please introduce explicit dot-product notation, for example, angle brackets, to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GCA-HNG is an empirical method evaluated on external retrieval benchmarks, with no prediction that reduces to its own inputs by construction.

full rationale

GCA-HNG is an empirical hard-negative generation method, not a derivation of benchmark numbers from its own assumptions. Its central mechanism (Section 3) constructs a mini-batch graph, propagates node and edge messages, and interpolates anchor-negative pairs via Eq. (6). The interpolation coefficients come from learned edge representations, and the hardness schedule eta = exp(-alpha/Javg) depends on the model's own average loss; this is a standard curriculum-style dependency rather than a fitted parameter renamed as a prediction, and the reported retrieval metrics are computed on held-out test sets against external baselines (Tables 1-2), so no benchmark result reduces by construction to the method's own outputs. The paper's self-citations in Related Works are contextual and not load-bearing for the main claim. The acknowledged limitation in Section 5, that correlations are confined to a mini-batch and hard negatives may not be synthesized for categories that are rarely co-sampled, is a scope restriction on the claimed 'global' perspective, not a circularity. Ablations comparing GCA-HNG with variants lacking global correlations provide independent empirical support for the mechanism. No equation-level or definition-level circularity was found.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a set of hand-set hyperparameters and on two domain assumptions: that synthetic interpolation preserves class semantics, and that batch-level correlations stand in for global correlations. The paper's empirical results are not derived from first principles, so these are the main epistemic prices.

free parameters (7)

similarity loss weight gamma_s = 1
Hand-set weight balancing Jsim in Jgen (Eq. 11); Table 5 ablates around it.
diversity loss weight gamma_d = 0.01 for PA, 0.03 for NP+
Hand-set weight balancing Jdiv; values differ by metric loss and were chosen via ablation on test sets.
pulling factor alpha = 5
Appears in eta = exp(-alpha/Javg) and in the Proxy Anchor loss; set manually in implementation details.
balancing factor beta = 2
Appears in gamma_n = exp(-beta/Jgen), controlling the proportion of synthetic negatives; set manually.
graph propagation steps K = 2 for CUB/Cars, 1 for SOP/InShop
Chosen from Figure 7 based on Recall@1 on test sets, not a held-out validation split.
attention heads H = 4 for CUB/Cars, 2 for SOP/InShop
Chosen from Figure 7 based on test performance; dataset-dependent.
sampling strategy N x m = 27x3 for CUB/Cars, 40x2 for SOP/InShop
Balanced sampling configuration selected from Section 4.5.3 tests on CUB and Cars196; fixed for SOP and InShop due to class size constraints.

assumptions (4)

domain assumption Interpolating between anchor and negative embeddings yields synthetic negatives whose class semantics are preserved enough for metric learning (Eq. 6 and Eq. 7).
The classification loss Jce trains a head on real samples but is applied to synthetic negatives; no guarantee is given that convex combinations in embedding space respect class boundaries.
domain assumption Mini-batch correlations with 27 or 40 sampled classes are a sufficient global context for hardness estimation.
The authors state in Section 5 that the framework only captures correlations within a mini-batch and may miss the categories most needing differentiation.
ad hoc to paper The dynamic scaling eta = exp(-alpha/Javg), with Javg from the last epoch, provides a useful annealing schedule for hardness.
No derivation is provided; this schedule is introduced to progressively increase hardness and is validated only empirically in Section 3.3.
domain assumption Using the test set as validation for hyperparameter tuning produces fair comparisons.
Section 4.1.1 explicitly says the testing set is treated as the validation set; this is a contested protocol, as cited in Musgrave et al. 2020, and can inflate results.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Globally Correlation-Aware Hard Negative Generation." pith.science (2026). https://pith.science/paper/LEO2GLXJ

@misc{pith2026241113145,
  author       = {Pith},
  title        = {Pith review of: Globally Correlation-Aware Hard Negative Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LEO2GLXJ}},
  note         = {Machine review of arXiv:2411.13145}
}

read the original abstract

Hard negative generation aims to generate informative negative samples that help to determine the decision boundaries and thus facilitate advancing deep metric learning. Current works select pair/triplet samples, learn their correlations, and fuse them to generate hard negatives. However, these works merely consider the local correlations of selected samples, ignoring global sample correlations that would provide more significant information to generate more informative negatives. In this work, we propose a Globally Correlation-Aware Hard Negative Generation (GCA-HNG) framework, which first learns sample correlations from a global perspective and exploits these correlations to guide generating hardness-adaptive and diverse negatives. Specifically, this approach begins by constructing a structured graph to model sample correlations, where each node represents a specific sample and each edge represents the correlations between corresponding samples. Then, we introduce an iterative graph message propagation to propagate the messages of node and edge through the whole graph and thus learn the sample correlations globally. Finally, with the guidance of the learned global correlations, we propose a channel-adaptive manner to combine an anchor and multiple negatives for HNG. Compared to current methods, GCA-HNG allows perceiving sample correlations with numerous negatives from a global and comprehensive perspective and generates the negatives with better hardness and diversity. Extensive experiment results demonstrate that the proposed GCA-HNG is superior to related methods on four image retrieval benchmark datasets. Codes and trained models are available at \url{https://github.com/PWenJay/GCA-HNG}.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

84 extracted references · 76 canonical work pages

[1]

In: CVPR, pp

Aziere, N., Todorovic, S.: Ensemble deep manifold similarity learning using hard proxies. In: CVPR, pp. 7299–7307 (2019)

2019
[2]

In: ECCV, pp

Bucher, M., Herbin, S., Jurie, F.: Hard negative mining for metric learning based zero-shot classification. In: ECCV, pp. 524–531. Springer (2016)

2016
[3]

In: ICCV, pp

Caron, M., Touvron, H., Misra, I., J´ egou, H., Mairal, J., Bojanowski, P., Joulin, A.: Emerging properties in self- supervised vision transformers. In: ICCV, pp. 9650–9660 (2021)

2021
[4]

IEEE Transactions on Pattern Analysis and Machine Intelligence 44(3), 1371–1384 (2022)

Chen, T., Lin, L., Chen, R., Hui, X., Wu, H.: Knowledge- guided multi-label few-shot learning for general image recog- nition. IEEE Transactions on Pattern Analysis and Machine Intelligence 44(3), 1371–1384 (2022)

2022
[5]

International Journal of Computer Vision 132, 6091–6106 (2024)

Chen, T., Pu, T., Liu, L., Shi, Y., Yang, Z., Lin, L.: Heteroge- neous semantic transfer for multi-label recognition with par- tial labels. International Journal of Computer Vision 132, 6091–6106 (2024)

work page 2024
[6]

IEEE Transac- tions on Pattern Analysis and Machine Intelligence 44(12), 9887–9903 (2022)

Chen, T., Pu, T., Wu, H., Xie, Y., Liu, L., Lin, L.: Cross- domain facial expression recognition: A unified evaluation benchmark and adversarial graph learning. IEEE Transac- tions on Pattern Analysis and Machine Intelligence 44(12), 9887–9903 (2022)

work page 2022
[7]

IEEE Transactions on Image Processing 33, 4811–4823 (2024)

Chen, T., Wang, W., Pu, T., Qin, J., Yang, Z., Liu, J., Lin, L.: Dynamic correlation learning and regularization for multi- label confidence calibration. IEEE Transactions on Image Processing 33, 4811–4823 (2024)

work page 2024
[8]

In: CVPR, pp

Chen, Z.M., Wei, X.S., Wang, P., Guo, Y.: Multi-label image recognition with graph convolutional networks. In: CVPR, pp. 5177–5186 (2019)

work page 2019

Show all 84 references

[9]

In: ECCV, pp

Dai, G., Zhang, Y., Ke, Q., Guo, Q., Huang, S.: One-dm: One-shot diffusion mimicker for handwritten text generation. In: ECCV, pp. 410–427 (2025)

2025
[10]

In: CVPR, pp

Dai, G., Zhang, Y., Wang, Q., Du, Q., Yu, Z., Liu, Z., Huang, S.: Disentangling writer and character styles for handwriting generation. In: CVPR, pp. 5977–5986 (2023) 19

2023
[11]

arXiv preprint arXiv:2010.11929 (2020)

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

2020 arXiv
[12]

IEEE Transactions on Image Processing 29, 2037–2051 (2019)

Duan, Y., Lu, J., Zheng, W., Zhou, J.: Deep adversarial met- ric learning. IEEE Transactions on Image Processing 29, 2037–2051 (2019)

2019
[13]

IEEE Transactions on Pattern Analysis and Machine Intelligence 45(2), 2505–2518 (2022)

Elezi, I., Seidenschwarz, J., Wagner, L., Vascon, S., Torci- novich, A., Pelillo, M., Leal-Taixe, L.: The group loss++: A deeper look into group loss for deep metric learning. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(2), 2505–2518 (2022)

2022
[14]

In: ECCV, pp

Elezi, I., Vascon, S., Torcinovich, A., Pelillo, M., Leal-Taix´ e, L.: The group loss for deep metric learning. In: ECCV, pp. 277–294. Springer (2020)

2020
[15]

In: CVPR, pp

Ermolov, A., Mirvakhabova, L., Khrulkov, V., Sebe, N., Os- eledets, I.: Hyperbolic vision transformers: Combining im- provements in metric learning. In: CVPR, pp. 7409–7419 (2022)

2022
[16]

Pattern Recognition 112, 107,795 (2021)

Gaji´ c, B., Amato, A., Gatta, C.: Fast hard negative mining for deep metric learning. Pattern Recognition 112, 107,795 (2021)

2021
[17]

In: AAAI, pp

Gu, G., Ko, B.: Symmetrical synthesis for deep metric learn- ing. In: AAAI, pp. 10,853–10,860 (2020)

2020
[18]

In: AAAI, pp

Gu, G., Ko, B., Kim, H.G.: Proxy synthesis: Learning with synthetic classes for deep metric learning. In: AAAI, pp. 1460–1468 (2021)

2021
[19]

In: CVPR, pp

Hadsell, R., Chopra, S., LeCun, Y.: Dimensionality reduction by learning an invariant mapping. In: CVPR, pp. 1735–1742 (2006)

2006
[20]

In: CVPR, pp

He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR, pp. 770–778 (2016)

2016
[21]

In: ACM MM, pp

Huang, H., Yang, D., Dai, G., Han, Z., Wang, Y., Lam, K.M., Yang, F., Huang, S., Liu, Y., He, M.: Agtgan: Unpaired image translation for photographic ancient character generation. In: ACM MM, pp. 5456–5467 (2022)

2022
[22]

In: KSEM, pp

Huang, J., Feng, Y., Zhou, M., Qiang, B.: Relationship-aware hard negative generation in deep metric learning. In: KSEM, pp. 388–400. Springer (2020)

2020
[23]

International Journal of Computer Vision 129, 1432–1450 (2021)

Husain, S.S., Ong, E.J., Bober, M.: Actnet: end-to-end learn- ing of feature activations and multi-stream aggregation for effective instance image retrieval. International Journal of Computer Vision 129, 1432–1450 (2021)

2021
[24]

In: ICML, pp

Ioffe, S., Szegedy, C.: Batch normalization: Accelerating deep network training by reducing internal covariate shift. In: ICML, pp. 448–456. pmlr (2015)

2015
[25]

In: ECCV, pp

Jin, S., RoyChowdhury, A., Jiang, H., Singh, A., Prasad, A., Chakraborty, D., Learned-Miller, E.: Unsupervised hard example mining from videos for improved object detection. In: ECCV, pp. 307–324 (2018)

2018
[26]

Kearnes, S., McCloskey, K., Berndl, M., Pande, V., Riley, P.: Molecular graph convolutions: moving beyond fingerprints. J. Comput. Aided Mol. Des 30, 595–608 (2016)

2016
[27]

In: CVPR, pp

Kim, S., Jeong, B., Kwak, S.: Hier: Metric learning beyond class labels via hierarchical regularization. In: CVPR, pp. 19,903–19,912 (2023)

2023
[28]

In: CVPR, pp

Kim, S., Kim, D., Cho, M., Kwak, S.: Proxy anchor loss for deep metric learning. In: CVPR, pp. 3238–3247 (2020)

2020
[29]

In: ICLR (2017)

Kipf, T.N., Welling, M.: Semi-supervised classification with graph convolutional networks. In: ICLR (2017)

2017
[30]

In: CVPR, pp

Ko, B., Gu, G.: Embedding expansion: Augmentation in em- bedding space for deep metric learning. In: CVPR, pp. 7255– 7264 (2020)

2020
[31]

In: ICCVW, pp

Krause, J., Stark, M., Deng, J., Fei-Fei, L.: 3d object repre- sentations for fine-grained categorization. In: ICCVW, pp. 554–561 (2013)

2013
[32]

In: IJCAI (2022)

Li, D., Wang, Z., Wang, J., Zhang, X., Ding, E., Wang, J., Zhang, Z.: Self-guided hard negative generation for unsuper- vised person re-identification. In: IJCAI (2022)

2022
[33]

Interna- tional Journal of Computer Vision 128, 2265–2278 (2020)

Li, Z., Tang, J., Zhang, L., Yang, J.: Weakly-supervised se- mantic guided hashing for social image retrieval. Interna- tional Journal of Computer Vision 128, 2265–2278 (2020)

2020
[34]

In: CVPR, pp

Liao, S., Shao, L.: Graph sampling based deep metric learn- ing for generalizable person re-identification. In: CVPR, pp. 7359–7368 (2022)

2022
[35]

In: CVPR, pp

Lim, J., Yun, S., Park, S., Choi, J.Y.: Hypergraph-induced semantic tuplet loss for deep metric learning. In: CVPR, pp. 212–222 (2022)

2022
[36]

Pattern Recog- nition Letters 186, 14–20 (2024)

Lin, Z., Li, J., Dai, G., Chen, T., Huang, S., Lin, J.: Con- trastive representation enhancement and learning for hand- written mathematical expression recognition. Pattern Recog- nition Letters 186, 14–20 (2024)

2024
[37]

Inter- national Journal of Computer Vision 128(8-9), 2223–2242 (2020)

Liu, H., Wang, R., Shan, S., Chen, X.: Learning multifunc- tional binary codes for personalized image retrieval. Inter- national Journal of Computer Vision 128(8-9), 2223–2242 (2020)

2020
[38]

In: CVPR, pp

Liu, Z., Luo, P., Qiu, S., Wang, X., Tang, X.: Deepfashion: Powering robust clothes recognition and retrieval with rich annotations. In: CVPR, pp. 1096–1104 (2016)

2016
[39]

In: ICLR (2018)

Loshchilov, I., Hutter, F.: Decoupled weight decay regular- ization. In: ICLR (2018)

2018
[40]

IEEE Transactions on Image Processing 26(9), 4269–4282 (2017)

Lu, J., Hu, J., Tan, Y.P.: Discriminative deep metric learning for face and kinship verification. IEEE Transactions on Image Processing 26(9), 4269–4282 (2017)

2017
[41]

In: ICCV, pp

Movshovitz-Attias, Y., Toshev, A., Leung, T.K., Ioffe, S., Singh, S.: No fuss distance metric learning using proxies. In: ICCV, pp. 360–368 (2017)

2017
[42]

In: ECCV, pp

Musgrave, K., Belongie, S., Lim, S.N.: A metric learning re- ality check. In: ECCV, pp. 681–699. Springer (2020)

2020
[43]

In: CVPR, pp

Oh Song, H., Xiang, Y., Jegelka, S., Savarese, S.: Deep metric learning via lifted structured feature embedding. In: CVPR, pp. 4004–4012 (2016)

2016
[44]

In: ICCV, pp

Qian, Q., Shang, L., Sun, B., Hu, J., Li, H., Jin, R.: Soft- triple loss: Deep metric learning without triplet sampling. In: ICCV, pp. 6450–6458 (2019)

2019
[45]

International Journal of Computer Vision pp

Rao, H., Leung, C., Miao, C.: Hierarchical skeleton meta- prototype contrastive learning with hard skeleton mining for unsupervised person re-identification. International Journal of Computer Vision pp. 1–23 (2023)

2023
[46]

In: CVPR, pp

Roth, K., Vinyals, O., Akata, Z.: Non-isotropy regularization for proxy-based deep metric learning. In: CVPR, pp. 7420– 7430 (2022)

2022
[47]

Inter- national Journal of Computer Vision 115(3), 211–252 (2015)

Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al.: Imagenet large scale visual recognition challenge. Inter- national Journal of Computer Vision 115(3), 211–252 (2015)

2015
[48]

In: CVPR, pp

Schroff, F., Kalenichenko, D., Philbin, J.: Facenet: A unified embedding for face recognition and clustering. In: CVPR, pp. 815–823 (2015)

2015
[49]

In: ICML, pp

Seidenschwarz, J.D., Elezi, I., Leal-Taix´ e, L.: Learning intra- batch connections for deep metric learning. In: ICML, pp. 9410–9421. PMLR (2021)

2021
[50]

In: CVPR, pp

Shrivastava, A., Gupta, A., Girshick, R.: Training region- based object detectors with online hard example mining. In: CVPR, pp. 761–769 (2016)

2016
[51]

In: ICCV, pp

Simo-Serra, E., Trulls, E., Ferraz, L., Kokkinos, I., Fua, P., Moreno-Noguer, F.: Discriminative learning of deep convo- lutional feature point descriptors. In: ICCV, pp. 118–126 (2015)

2015
[52]

In: CVPR W, pp

Smirnov, E., Melnikov, A., Oleinik, A., Ivanova, E., Kali- novskiy, I., Luckyanets, E.: Hard example mining with aux- iliary embeddings. In: CVPR W, pp. 37–46 (2018) 20

2018
[53]

In: NeurIPS, pp

Sohn, K.: Improved deep metric learning with multi-class n- pair loss objective. In: NeurIPS, pp. 1857–1865 (2016)

2016
[54]

In: CVPR, pp

Suh, Y., Han, B., Kim, W., Lee, K.M.: Stochastic class-based hard example mining for deep metric learning. In: CVPR, pp. 7251–7259 (2019)

2019
[55]

In: CVPR, pp

Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., Rabinovich, A.: Go- ing deeper with convolutions. In: CVPR, pp. 1–9 (2015)

2015
[56]

spot face recognition

Tan, Z., Liu, A., Wan, J., Liu, H., Lei, Z., Guo, G., Li, S.Z.: Cross-batch hard example mining with pseudo large batch for id vs. spot face recognition. IEEE Transactions on Image Processing 31, 3224–3235 (2022)

2022
[57]

In: ECCV, pp

Teh, E.W., DeVries, T., Taylor, G.W.: Proxynca++: Revisit- ing and revitalizing proxy neighborhood component analysis. In: ECCV, pp. 448–464. Springer (2020)

2020
[58]

In: ICML, pp

Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., J´ egou, H.: Training data-efficient image transformers & distillation through attention. In: ICML, pp. 10,347–10,357. PMLR (2021)

2021
[59]

NeurIPS 30 (2017)

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. NeurIPS 30 (2017)

2017
[60]

In: ICLR (2018)

Veliˇ ckovi´ c, P., Cucurull, G., Casanova, A., Romero, A., Li` o, P., Bengio, Y.: Graph attention networks. In: ICLR (2018)

2018
[61]

In: ICLR, pp

Venkataramanan, S., Psomas, B., Kijak, E., Amsaleg, L., Karantzalos, K., Avrithis, Y.: It takes two to tango: Mixup for deep metric learning. In: ICLR, pp. 1–21 (2022)

2022
[62]

In: CVPR, pp

Wang, C., Zheng, W., Li, J., Zhou, J., Lu, J.: Deep factorized metric learning. In: CVPR, pp. 7672–7682 (2023)

2023
[63]

In: ICCV, pp

Wang, J., Zhou, F., Wen, S., Liu, X., Lin, Y.: Deep metric learning with angular loss. In: ICCV, pp. 2593–2601 (2017)

2017
[64]

In: CVPR, pp

Wang, X., Han, X., Huang, W., Dong, D., Scott, M.R.: Multi- similarity loss with general pair weighting for deep metric learning. In: CVPR, pp. 5022–5030 (2019)

2019
[65]

In: AAAI, vol

Wang, Y., He, D., Li, F., Long, X., Zhou, Z., Ma, J., Wen, S.: Multi-label classification with label graph superimposing. In: AAAI, vol. 34, pp. 12,265–12,272 (2020)

2020
[66]

Journal of Ma- chine Learning Research 10(2) (2009)

Weinberger, K.Q., Saul, L.K.: Distance metric learning for large margin nearest neighbor classification. Journal of Ma- chine Learning Research 10(2) (2009)

2009
[67]

California Institute of Technology (2010)

Welinder, P., Branson, S., Mita, T., Wah, C., Schroff, F., Belongie, S., Perona, P.: Caltech-ucsd birds 200. California Institute of Technology (2010)

2010
[68]

In: W ACV, pp

Xuan, H., Stylianou, A., Pless, R.: Improved embeddings with easy positive triplet mining. In: W ACV, pp. 2474–2482 (2020)

2020
[69]

In: ICCV, pp

Yang, B., Sun, H., Li, F.W., Chen, Z., Cai, J., Song, C.: Hse: Hybrid species embedding for deep metric learning. In: ICCV, pp. 11,047–11,057 (2023)

2023
[70]

In: CVPR, pp

Yang, L., Zhan, X., Chen, D., Yan, J., Loy, C.C., Lin, D.: Learning to cluster faces on an affinity graph. In: CVPR, pp. 2298–2306 (2019)

2019
[71]

In: W ACV, pp

Yang, Z., Bastan, M., Zhu, X., Gray, D., Samaras, D.: Hierar- chical proxy-based loss for deep metric learning. In: W ACV, pp. 1859–1868 (2022)

2022
[72]

In- ternational Journal of Computer Vision 129, 484–500 (2021)

Yu, Q., Song, J., Song, Y.Z., Xiang, T., Hospedales, T.M.: Fine-grained instance-level sketch-based image retrieval. In- ternational Journal of Computer Vision 129, 484–500 (2021)

2021
[73]

IEEE Transactions on Neural Networks and Learning Systems 34(12), 10,528– 10,537 (2022)

Zeng, Y., Wang, Y., Liao, D., Li, G., Huang, W., Xu, J., Cao, D., Man, H.: Keyword-based diverse image retrieval with variational multiple instance graph. IEEE Transactions on Neural Networks and Learning Systems 34(12), 10,528– 10,537 (2022)

2022
[74]

arXiv preprint arXiv:1811.12649 (2018)

Zhai, A., Wu, H.Y.: Classification is a strong baseline for deep metric learning. arXiv preprint arXiv:1811.12649 (2018)

2018 arXiv
[75]

In: CVPR, pp

Zhang, B., Zheng, W., Zhou, J., Lu, J.: Attributable visual similarity learning. In: CVPR, pp. 7532–7541 (2022)

2022
[76]

In: AAAI, vol

Zhang, C., Luo, L., Gu, B.: Denoising multi-similarity for- mulation: a self-paced curriculum-driven approach for robust metric learning. In: AAAI, vol. 37, pp. 11,183–11,191 (2023)

2023
[77]

In: ECCV, pp

Zhao, Y., Jin, Z., Qi, G.j., Lu, H., Hua, X.s.: An adversarial approach to hard triplet generation. In: ECCV, pp. 501–517 (2018)

2018
[78]

In: CVPR, pp

Zheng, W., Chen, Z., Lu, J., Zhou, J.: Hardness-aware deep metric learning. In: CVPR, pp. 72–81 (2019)

2019
[79]

IEEE Transactions on Pattern Analysis and Ma- chine Intelligence 43(9), 3214–3228 (2021)

Zheng, W., Lu, J., Zhou, J.: Hardness-aware deep metric learning. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence 43(9), 3214–3228 (2021)

2021
[80]

Neurocomputing 498, 59–74 (2022)

Zhu, C., Hu, Z., Dong, H., He, G., Yu, Z., Zhang, S.: Con- struct informative triplet with two-stage hard-sample gener- ation. Neurocomputing 498, 59–74 (2022)

2022
[81]

International Journal of Com- puter Vision 131(11), 2959–2976 (2023)

Zhu, J., Liu, L., Zhan, Y., Zhu, X., Zeng, H., Tao, D.: Attribute-image person re-identification via modal- consistent metric learning. International Journal of Com- puter Vision 131(11), 2959–2976 (2023)

2023
[82]

IEEE Transactions on Image Processing 30, 7593–7607 (2021)

Zhu, S., Yang, T., Chen, C.: Visual explanation for deep met- ric learning. IEEE Transactions on Image Processing 30, 7593–7607 (2021)

2021
[83]

Pattern Recognition 95, 211–222 (2019)

Zhu, X., Jing, X.Y., Zhang, F., Zhang, X., You, X., Cui, X.: Distance learning by mining hard and easy negative samples for person re-identification. Pattern Recognition 95, 211–222 (2019)

2019
[84]

In: NeurIPS, pp

Zhu, Y., Yang, M., Deng, C., Liu, W.: Fewer is more: a deep graph metric learning perspective using fewer proxies. In: NeurIPS, pp. 17,792–17,803 (2020)

2020

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

In: CVPR, pp

Aziere, N., Todorovic, S.: Ensemble deep manifold similarity learning using hard proxies. In: CVPR, pp. 7299–7307 (2019)

2019

[2] [2]

In: ECCV, pp

Bucher, M., Herbin, S., Jurie, F.: Hard negative mining for metric learning based zero-shot classification. In: ECCV, pp. 524–531. Springer (2016)

2016

[3] [3]

In: ICCV, pp

Caron, M., Touvron, H., Misra, I., J´ egou, H., Mairal, J., Bojanowski, P., Joulin, A.: Emerging properties in self- supervised vision transformers. In: ICCV, pp. 9650–9660 (2021)

2021

[4] [4]

IEEE Transactions on Pattern Analysis and Machine Intelligence 44(3), 1371–1384 (2022)

Chen, T., Lin, L., Chen, R., Hui, X., Wu, H.: Knowledge- guided multi-label few-shot learning for general image recog- nition. IEEE Transactions on Pattern Analysis and Machine Intelligence 44(3), 1371–1384 (2022)

2022

[5] [5]

International Journal of Computer Vision 132, 6091–6106 (2024)

Chen, T., Pu, T., Liu, L., Shi, Y., Yang, Z., Lin, L.: Heteroge- neous semantic transfer for multi-label recognition with par- tial labels. International Journal of Computer Vision 132, 6091–6106 (2024)

work page 2024

[6] [6]

IEEE Transac- tions on Pattern Analysis and Machine Intelligence 44(12), 9887–9903 (2022)

Chen, T., Pu, T., Wu, H., Xie, Y., Liu, L., Lin, L.: Cross- domain facial expression recognition: A unified evaluation benchmark and adversarial graph learning. IEEE Transac- tions on Pattern Analysis and Machine Intelligence 44(12), 9887–9903 (2022)

work page 2022

[7] [7]

IEEE Transactions on Image Processing 33, 4811–4823 (2024)

Chen, T., Wang, W., Pu, T., Qin, J., Yang, Z., Liu, J., Lin, L.: Dynamic correlation learning and regularization for multi- label confidence calibration. IEEE Transactions on Image Processing 33, 4811–4823 (2024)

work page 2024

[8] [8]

In: CVPR, pp

Chen, Z.M., Wei, X.S., Wang, P., Guo, Y.: Multi-label image recognition with graph convolutional networks. In: CVPR, pp. 5177–5186 (2019)

work page 2019