Pith. sign in

REVIEW 4 major objections 4 minor 79 references

Locality Preserving Markovian Transition for Instance Retrieval

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes LPMT, a manifold-aware distance measured as the minimum transition flow cost between per-image probability distributions, and reports consistent gains in instance retrieval re-ranking.

desk verdict Strong empirical re-ranking results, but the thermodynamic multi-state story is untested as written; ship to review with major revision. read the letter →

arxiv 2506.05196 v1 pith:ZYM6VGP4 submitted 2025-06-05 cs.LG

Jifei Luo , Wenzheng Wu , Hantao Yao , Lu Yu , Changsheng Xu This is my paper · ORCID
classification cs.LG
keywords instanceretrievalre-rankingmanifoldrankingdiffusionoptimaltransportWassersteindistancethermodynamictransitionlocalitypreserving
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

Diffusion-based re-ranking propagates similarity along a k-nearest-neighbour graph, but positive signal fades over several steps, so distant correct matches lose discriminative power. The paper proposes LPMT, which addresses this by encoding each instance as a probability distribution over its reciprocal local neighbourhood and then measuring the distance between two instances as the minimum cost of a multi-stage thermodynamic transition that moves one distribution to the other through intermediate local states. This transition cost is blended with the Euclidean feature distance to re-rank the initial retrieval results. Experiments on the revised Oxford and Paris landmark benchmarks (ROxf, RPar) and their 1M-distractor versions, using features from R-GeM, DOLG, CVNet, MAC, R-MAC, DELG, and SENet, report mAP improvements over diffusion, query expansion, and learning-based baselines. The motivation is that such a manifold-aware distance preserves local effectiveness while capturing global manifold structure, which is what the paper seeks to establish.

What carries the argument

The central object is the Thermodynamic Markovian Transition (TMT), whose defining identity is the equivalence between the minimum transition-flow cost of a master-equation Markov process and a path-sum of Wasserstein distances, $d'(i,j) = \min_{\pi} \sum_k W_1(p_{i_k}, p_{i_{k+1}})$. That identity connects stochastic thermodynamics to optimal transport and makes the distance computable via entropy-regularized Sinkhorn iterations. The machinery also includes Bidirectional Collaborative Diffusion (BCD), which jointly optimizes aggregation weights and a bidirectional diffusion objective over an enlarged set of k-NN graphs to produce a robust similarity matrix, and Locality State Embedding (LSE), which converts each row of that matrix into a probability distribution over reciprocal neighbours. Together these components turn raw deep features into manifold-aware state distributions and a distance that can be used for re-ranking.

What would settle it

Compute $d'(i,j)$ from Eq. (20) with one intermediate state and with many for a sample of image pairs; since $W_1$ is a metric, the values coincide, which would show the multi-state transition contributes nothing and any gains come from LSE, BCD, or the Euclidean blend. Alternatively, check the released implementation to see whether $W_1$ is restricted to transitions along local graph edges; if it is not, the computed distance reduces to the direct earth-mover distance.

Watch

Extended reading notes

Core claim

The central claim is that a long-term thermodynamic Markovian transition over locally embedded probability distributions provides an effective manifold distance for instance retrieval. In the framework, each instance is represented by a sparse probability distribution built from the smoothed similarity matrix produced by Bidirectional Collaborative Diffusion and a k-reciprocal local region. The distance between two instances is then defined by the Thermodynamic Markovian Transition cost $d'(i,j) = \min_{\pi} \sum_{k=0}^{K-1} W_1(p_{i_k}, p_{i_{k+1}})$, where $W_1$ is the earth-mover (Wasserstein) distance between consecutive state distributions; the paper proves in Appendix B.2 that this equals the minimum flow cost of a continuous-time Markov process governed by the master equation $\dot{q}_t = T_t q_t$. The final retrieval distance is $d^*(i,j) = \theta d(i,j) + (1-\theta)d'(i,j)$, balancing Euclidean and manifold distances. The paper reports that this distance yields higher mAP than prior re-ranking methods across multiple feature extractors and benchmark protocols.

Load-bearing premise

The load-bearing premise is that the per-step cost in the multi-state path is not a true metric, so the intermediate states can change the computed distance; the earth-mover distance is a metric, so as written the minimum over paths collapses to the direct earth-mover distance and the thermodynamic stages have no effect.

Editorial extensions

If this is right

  • Re-ranking with LPMT improves mAP over query-expansion, diffusion-based, context-based, and learning-based baselines on ROxf, RPar, and their 1M-distractor versions across R-GeM, DOLG, CVNet, MAC, R-MAC, DELG, and SENet features.
  • Because LSE only uses reciprocal local neighbours, the distance is more robust to erroneous or missing k-NN graph connections than standard diffusion.
  • LPMT transfers to unsupervised content-based image retrieval, improving mAP and Recall@1 on CUB200, Indoor, and Caltech101.
  • Restricting re-ranking to the top-k images reduces the O(n^3) computation to O(k^3), keeping runtime under a few seconds for k=5000.
  • Ablations show the TMT distance beats cosine, Euclidean, Jaccard, and total-variation distances between the same embeddings, indicating the transition cost itself carries the improvement.

Reading between the lines

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

  • If the metric-collapse observation holds, the reported gains are likely attributable to BCD and LSE rather than to the multi-state thermodynamic stage; a locally graph-constrained transport cost would be a natural variant that could make the intermediate states genuinely change the optimum.
  • The distance formulation is feature-agnostic and could be applied to other retrieval settings, including text or multimodal embeddings, wherever a Euclidean baseline exists.
  • The entropy-regularized Sinkhorn solver introduces a temperature-like parameter; varying it could reveal whether the benefit of the transition cost comes from smoothing rather than from manifold geometry.
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.

Referee Report

4 major / 4 minor

Summary. The paper proposes LPMT, a re-ranking framework for instance retrieval comprising three components: Bidirectional Collaborative Diffusion (BCD) to fuse multiple affinity graphs, Locality State Embedding (LSE) to represent each image as a probability distribution over its local neighborhood, and Thermodynamic Markovian Transition (TMT) to compute a distance between these distributions via a multi-state transition flow. The final distance is a weighted combination of the original Euclidean distance and the TMT cost. Experiments on ROxf/RPar and on content-based retrieval datasets report consistent mAP improvements over a wide range of baselines, including substantial gains over the previous state of the art (e.g., R-GeM ROxf medium 84.7 vs. CAS 80.7).

Significance. If the multi-state thermodynamic transition were the source of the reported gains, this would be a meaningful contribution to manifold-based re-ranking: the empirical improvements are large, consistent across six feature extractors, and the paper includes a broad set of baselines. The paper also provides a formal connection between the flow cost and the W1 distance. However, the central claim about the multi-state path is not operationalized: the paper never defines the feasible path set, gives no algorithm for the constrained path optimization, and the TMT ablation omits the direct W1 baseline. As presented, the evidence does not isolate the contribution of the thermodynamic path mechanism, which is the paper's main novelty.

major comments (4)
  1. [§4.3, Eq. (20)] The paper never specifies the set of feasible transition paths π: it states that each state must be reachable from its predecessor 'within the local region', but it does not define what 'local region' means for the distributions {p_i} or how the constrained shortest path in Eq. (20) is computed. Appendix B.3 only provides a Sinkhorn iteration for pairwise W1. Table 6 compares TMT against Cosine, Euclidean, Jaccard, and Total Variation distances, but omits the direct W1 baseline (K=0 or a single-edge path) and an unconstrained-path baseline. As a result, the reported gains do not establish that the multi-state transition contributes; they could be entirely due to the LSE plus W1 pipeline. This is the load-bearing claim of the paper and must be supported by an explicit path-optimization algorithm and the missing baselines.
  2. [Appendix B.2, proof of Eq. (B.8)] The proof shows only that for a fixed path π, the minimum flow cost over transition rates equals the sum of W1 distances along the path; it does not show that the minimum over paths is different from the direct distance W1(p_i, p_j), nor does it enforce any local restriction on the transport plan. In the constructive part (Eqs. B.21–B.28), mass is moved directly between arbitrary vertices r_m and s_m over edges with cost d(r_m, s_m), which is exactly the unconstrained optimal transport problem. If the intended 'local region' constraint applies to these transport edges, the proof does not model it; if it applies only to the sequence of intermediate states, that constraint and its effect on the optimum need to be stated formally.
  3. [§4.1, Eq. (7)] The gradient of J in Eq. (7) is inconsistent with the quadratic form in Eq. (6) and with the correct derivative in Appendix A.1 (Eq. A.9). For symmetric ar{S}^v, the derivative of vec(F)^ op (I - ar{S}^v) vec(F) is 2(I - ar{S}^v)vec(F), not (2I - ar{S}^v)vec(F). The closed-form solution in Eq. (8) corresponds to the correct gradient, so this appears to be a typographical error, but it should be corrected because the text presents Eq. (7) as the basis for Eq. (8).
  4. [§4.3, text after Eq. (18)] The statement that 'an additional power term can be applied to the distance in Eq. (18)' is never formalized; no definition, default value, or experimental setting is given. Either specify the term (e.g., d(r,s)^p) and its value in the experiments, or remove the sentence, because the reader cannot determine what was actually evaluated.
minor comments (4)
  1. [§4.3, Eq. (18) and surrounding text] 'Temperate states' appears to be a typo for 'temporary states' or 'intermediate states'.
  2. [Table 5] The header 'k-reciprcal' should be 'k-reciprocal'.
  3. [Appendix A.3, Algorithm 2] The initialization of P(0) and R(0) is stated in a single line, but R(0) is not subsequently used in the loop; clarify the notation and the stopping criterion.
  4. [§5.1, Implementation Details] The paper does not report the TMT-specific hyperparameters (e.g., τ, K, the Sinkhorn regularization ε, and the number of path states), even though the sensitivity analysis in Fig. 5 covers only k1, k2, κ, θ, μ, and σ. Please add these details to make the experiments reproducible.

Circularity Check

2 steps flagged · score 5.0 of 10

The final distance is a fitted blend of Euclidean and W1-based terms, and the TMT component is, by the paper's own reduction, a relabeled W1 distance that is never compared against plain W1; the derivation itself is mostly self-contained.

  1. fitted input called prediction [Section 5.3, 'Sensitivity of Hyper-parameters', Fig. 5(d), Eq. (21), Tables 1-3]
    "Fig. 5(d) reveals that θ = 0.5 yields the optimal result, demonstrating that incorporating the original distance enhances the retrieval robustness."

    The final distance is d∗(i,j)=θd(i,j)+(1−θ)d′(i,j) (Eq. 21), and the ROxf/RPar columns of Tables 1-3 are reported as the method's predictive performance. However, θ (and k1, k2, κ, μ, σ in the same figure) are selected by scanning mAP on the ROxf medium and hard protocols, which are the very test protocols whose scores are then tabulated. The headline ROxf numbers are therefore the maxima of the tuning curves, not independent predictions of the method; presenting them as evidence that the thermodynamic distance is effective makes the headline evaluation circular with the fitted hyperparameters.

  2. renaming known result [Section 4.3, Eq. (20); Appendix B.2-B.3]
    "under the assumption that each transition only takes place in local regions, the Wasserstein distance W1 can serve as a valid equivalency, followed by: d′(i,j)=minπ Σ_{k=0}^{K−1} W1(p_ik,p_ik+1)."

    By the paper's own Eq. (20), the 'Thermodynamic Markovian Transition' cost is defined as a shortest path over W1 distances, where W1 is the standard optimal-transport distance defined in Eq. (B.2). Appendix B.3 provides only the Sinkhorn fixed-point solver for pairwise W1(p_start,p_end); no algorithm is given for the constrained path minimization over intermediate states, and Table 6 compares TMT against total variation and Jaccard but never against direct W1. Thus, unless the missing path solver changes the value, the computed quantity is exactly the W1 distance between LSE embeddings; the multi-state thermodynamic framing is a relabeling of a known metric, and the claimed 'long-term transition' mechanism is not demonstrated to contribute anything beyond W1.

full rationale

The paper's derivation chain (BCD → LSE → TMT) is largely self-contained: BCD has a closed-form iterative solution (Appendix A), LSE is a defined neighborhood-weighted embedding (Eqs. 14-16), and the TMT-to-W1 equivalency is proved internally in Appendix B rather than imported from a citation. The self-citation to Luo et al. (2024) for the Bidirectional Similarity Diffusion objective is not load-bearing for the central claim, because BCD is ablated against BSD in Fig. 4(a) and the TMT/LSE contributions have independent derivations. However, two partial circularities remain. First, the headline ROxf/RPar results are produced after tuning the hyperparameters (θ, k1, k2, κ, μ, σ) on those same test protocols, so the reported mAP values are fitted rather than predicted on the primary benchmark. Second, the paper's own theorem reduces the thermodynamic transition cost to a shortest-path sum of W1 distances, while the implementation description and ablations never show that a multi-state constrained path is actually solved or that it differs from direct W1; the novelty claim therefore reduces to a relabeled W1 distance unless the missing path optimization is supplied. These issues make the central effectiveness claim partially circular, though the core derivation is not a tautological reduction.

Assumptions & free parameters 9 free parameters · 4 assumptions · 2 invented entities

The central method relies on a long list of tuned constants and on the equivalence between thermodynamic flow and W1. The most consequential item is the ad hoc equivalence in Eq. (20), since it makes the multi-state mechanism vacuous. Many constants such as mu, sigma, lambda, Sinkhorn epsilon, tau, and K are not reported.

free parameters (9)
  • k1 = 60
    Local region size in LSE Eq. (15), chosen by maximizing mAP in Fig. 5(a) on ROxf medium and hard.
  • k2 = 7
    Confident neighbor count in Eq. (16), tuned in Fig. 5(b).
  • kappa = 2
    Reciprocal neighbor emphasis weight in Eq. (16), tuned in Fig. 5(c).
  • theta = 0.5
    Balance between Euclidean distance and transition cost in Eq. (21), tuned in Fig. 5(d).
  • mu = not reported
    BCD regularization weight in Eq. (3) and Eq. (5); sensitivity is shown in Fig. 5(e), but the final value is not stated.
  • sigma = not reported
    Affinity bandwidth in Eq. (2); sensitivity is shown in Fig. 5(f), but the final value is not stated.
  • lambda = not reported
    BCD weight regularization in Eq. (3); no value or sensitivity analysis is given.
  • graph scale factors = [1/sqrt(2), 1, sqrt(2)]
    Graph set construction for BCD in Sec. 5.3, chosen by hand.
  • Sinkhorn epsilon, tau, path length K = not reported
    Entropy regularization in Eq. (B.30) and transition interval or path length in Eq. (B.3); none are specified or ablated.
assumptions (4)
  • domain assumption The k-NN affinity graph W in Eq. (2) approximates the low-dimensional data manifold.
    Invoked in Sec. 3 and throughout BCD and TMT; if graph edges are wrong, propagated similarities are wrong.
  • domain assumption Local neighborhoods defined by k-reciprocal nearest neighbors contain mostly same-category instances.
    Used in LSE Eq. (14)-(16) to build locality state distributions; no label-based verification is provided.
  • domain assumption Euclidean distance in the original feature space is the correct ground cost for the transition and Wasserstein distance.
    Eq. (18) and Eq. (B.2) use d(r,s) from Eq. (1) as the edge cost; the manifold claim is not otherwise grounded.
  • ad hoc to paper The minimum thermodynamic flow cost per stage equals unconstrained W1 between locality embeddings, and the multi-state path has an effect.
    Eq. (20) and Appendix B.2 equate each stage with W1, but the proof does not enforce the local-edge constraint used to motivate intermediate states; by triangle inequality the path minimum collapses to direct W1.
invented entities (2)
  • Locality state distribution p_i independent evidence
    purpose: Represents each instance as a probability distribution over the manifold for distance computation in Eq. (16).
    The representation is used in all experiments and can be evaluated on held-out retrieval benchmarks, so it has a falsifiable handle, but it is a learned embedding, not a physical entity.
  • Thermodynamic Markovian transition flow q_t
    purpose: Conceptual multi-state process connecting two instance distributions; its minimum cost is claimed as the retrieval distance in Eq. (18).
    The flow is never constructed in the reported implementation; Eq. (20) replaces it with W1, so the entity has no independent observable handle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Locality Preserving Markovian Transition for Instance Retrieval." pith.science (2026). https://pith.science/paper/ZYM6VGP4

@misc{pith2026250605196,
  author       = {Pith},
  title        = {Pith review of: Locality Preserving Markovian Transition for Instance Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZYM6VGP4}},
  note         = {Machine review of arXiv:2506.05196}
}
read the original abstract

Diffusion-based re-ranking methods are effective in modeling the data manifolds through similarity propagation in affinity graphs. However, positive signals tend to diminish over several steps away from the source, reducing discriminative power beyond local regions. To address this issue, we introduce the Locality Preserving Markovian Transition (LPMT) framework, which employs a long-term thermodynamic transition process with multiple states for accurate manifold distance measurement. The proposed LPMT first integrates diffusion processes across separate graphs using Bidirectional Collaborative Diffusion (BCD) to establish strong similarity relationships. Afterwards, Locality State Embedding (LSE) encodes each instance into a distribution for enhanced local consistency. These distributions are interconnected via the Thermodynamic Markovian Transition (TMT) process, enabling efficient global retrieval while maintaining local effectiveness. Experimental results across diverse tasks confirm the effectiveness of LPMT for instance retrieval.

Figures

Figures reproduced from arXiv: 2506.05196 by the authors.

Figure 1
Figure 1. Illustration of Locality Preserving Markovian Transition. Each instance is embedded as a distribution within the manifold, with its characteristics shaped by the intrinsic local neighborhood structure. Distant distributions are bridged via multiple intermedi￾ate states, where each transition is confined to a local region and governed by the master equation. The minimum transition cost then serves as an effective dis… view at source ↗
Figure 2
Figure 2. Comparison of ranking results based on (a) Euclidean distance and (b) manifold-aware distance in the feature space. instance into a manifold-aware space to perform re-ranking, where similar images exhibit higher contextual consistency. Learning-based Methods. Recently, deep learning methods have also been introduced to assist with re-ranking. For example, Gordo et al. (2020) and Ouyang et al. (2021) lever￾age the ro… view at source ↗
Figure 3
Figure 3. An illustration of the proposed Bidirectional Collabora￾tive Diffusion (BCD) algorithm with three scales, where different connection strategies are employed to accommodate diverse data distributions. BCD automatically performs diffusion and integra￾tion to generate a robust similarity matrix. The optimization problem in Eq. (3), which simultaneously depends on β and F, is inherently complex and imprac￾tical for a di… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Ablations of BCD. (a) Performance comparison with BSD. (b) Convergence analysis of BCD towards the target matrix. 5.2. Main Results Comparison of Instance Retrieval. As summarized in Ta￾ble 1, we evaluate our proposed LPMT against a wide range of re-ranking approaches,…
Figure 5
Figure 5. Figure 5: Sensitivity analysis of hyper-parameters based on image features extracted by R-GeM. (a) Effect of k1. (b) Effect of k2. (c) Effect of κ. (d) Effect of θ. (e) Effect of µ. (f) Effect of σ. 6.7%/6.2%/5.9%/5.2% and 7.8%/7.1%/7.2%/6.1% based on R-GeM. This suggests that t…
Figure 6
Figure 6. Figure 6: This figure showcases the quantitative evaluation of our proposed LPMT in comparison with retrieval results based on Euclidean distance. The region of interest in the query image is highlighted by an orange bounding box on the left. On the right, we visualize the retri…
Figure 7
Figure 7. Figure 7: (a) The t-SNE visualization of the original image features directly extracted by the deep retrieval model. (b) The t-SNE visualization of the distributions produced by the Locality State Embedding (LSE) strategy using cosine similarity weights. Compared to the original…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 76 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    and Zisserman, A

    Arandjelovi \'c , R. and Zisserman, A. Three things everyone should know to improve object retrieval. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012

    work page 2012

  3. [3]

    Self- RAG : Learning to retrieve, generate, and critique through self-reflection

    Asai, A., Wu, Z., Wang, Y., Sil, A., and Hajishirzi, H. Self- RAG : Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  4. [4]

    and Bai, X

    Bai, S. and Bai, X. Sparse contextual activation for efficient visual re-ranking. IEEE Transactions on Image Processing, 25 0 (3): 0 1056--1069, 2016

    work page 2016

  5. [5]

    Bai, S., Bai, X., Tian, Q., and Latecki, L. J. Regularized diffusion process for visual retrieval. In Proceedings of the AAAI Conference on Artificial Intelligence, 2017 a

    work page 2017

  6. [6]

    J., and Tian, Q

    Bai, S., Zhou, Z., Wang, J., Bai, X., Latecki, L. J., and Tian, Q. Ensemble diffusion for retrieval. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017 b

    work page 2017

  7. [7]

    Bai, S., Bai, X., Tian, Q., and Latecki, L. J. Regularized diffusion process on bidirectional context for object retrieval. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41 0 (5): 0 1213--1226, 2019 a

    work page 2019

  8. [8]

    H., and Latecki, L

    Bai, S., Tang, P., Torr, P. H., and Latecki, L. J. Re-ranking via metric fusion for object retrieval and person re-identification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019 b

    work page 2019

Show all 79 references
  1. [9]

    J., and Tian, Q

    Bai, S., Zhou, Z., Wang, J., Bai, X., Latecki, L. J., and Tian, Q. Automatic ensemble diffusion for 3d shape and image retrieval. IEEE Transactions on Image Processing, 28 0 (1): 0 88--101, 2019 c

    2019

  2. [10]

    Barato, A. C. and Seifert, U. Thermodynamic uncertainty relation for biomolecular processes. Phys. Rev. Lett., 114: 0 158101, Apr 2015

    2015

  3. [11]

    Borgeaud, S., Mensch, A., Hoffmann, J., Cai, T., Rutherford, E., Millican, K., Van Den Driessche, G. B., Lespiau, J.-B., Damoc, B., Clark, A., De Las Casas, D., Guy, A., Menick, J., Ring, R., Hennigan, T., Huang, S., Maggiore, L., Jones, C., Cassirer, A., Brock, A., Paganini, ...

    2022

  4. [12]

    Unifying deep local and global features for image search

    Cao, B., Araujo, A., and Sim, J. Unifying deep local and global features for image search. In European Conference on Computer Vision (ECCV), 2020

    2020

  5. [13]

    Explore-exploit graph traversal for image retrieval

    Chang, C., Yu, G., Liu, C., and Volkovs, M. Explore-exploit graph traversal for image retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  6. [14]

    Benchmarking large language models in retrieval-augmented generation

    Chen, J., Lin, H., Han, X., and Sun, L. Benchmarking large language models in retrieval-augmented generation. Proceedings of the AAAI Conference on Artificial Intelligence, 2024

    2024

  7. [15]

    Total recall: Automatic query expansion with a generative feature model for object retrieval

    Chum, O., Philbin, J., Sivic, J., Isard, M., and Zisserman, A. Total recall: Automatic query expansion with a generative feature model for object retrieval. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2007

    2007

  8. [16]

    Sinkd: Sinkhorn distance minimization for knowledge distillation

    Cui, X., Qin, Y., Gao, Y., Zhang, E., Xu, Z., Wu, T., Li, K., Sun, X., Zhou, W., and Li, H. Sinkd: Sinkhorn distance minimization for knowledge distillation. IEEE Transactions on Neural Networks and Learning Systems, 2024 a

    2024

  9. [17]

    Sinkhorn distance minimization for knowledge distillation

    Cui, X., Qin, Y., Gao, Y., Zhang, E., Xu, Z., Wu, T., Li, K., Sun, X., Zhou, W., and Li, H. Sinkhorn distance minimization for knowledge distillation. In LREC-COLING, pp.\ 14846--14858, 2024 b

    2024

  10. [18]

    Optical: Leveraging optimal transport for contribution allocation in dataset distillation

    Cui, X., Qin, Y., Xie, L., Zhou, W., Li, H., and Li, H. Optical: Leveraging optimal transport for contribution allocation in dataset distillation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025 a

    2025

  11. [19]

    Layoutenc: Leveraging enhanced layout representations for transformer-based complex scene synthesis

    Cui, X., Sun, Q., Wang, M., Li, L., Zhou, W., and Li, H. Layoutenc: Leveraging enhanced layout representations for transformer-based complex scene synthesis. ACM Transactions on Multimedia Computing, Communications and Applications, 2025 b

    2025

  12. [20]

    Multi-level optimal transport for universal cross-tokenizer knowledge distillation on language models

    Cui, X., Zhu, M., Qin, Y., Xie, L., Zhou, W., and Li, H. Multi-level optimal transport for universal cross-tokenizer knowledge distillation on language models. Proceedings of the AAAI Conference on Artificial Intelligence, 2025 c

    2025

  13. [21]

    Sinkhorn distances: Lightspeed computation of optimal transport

    Cuturi, M. Sinkhorn distances: Lightspeed computation of optimal transport. In Advances in Neural Information Processing Systems, 2013

    2013

  14. [22]

    and Doucet, A

    Cuturi, M. and Doucet, A. Fast computation of wasserstein barycenters. In Xing, E. P. and Jebara, T. (eds.), Proceedings of the 31st International Conference on Machine Learning, Proceedings of Machine Learning Research, pp.\ 685--693. PMLR, 2014

    2014

  15. [23]

    BERT : Pre-training of deep bidirectional transformers for language understanding

    Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. BERT : Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technol...

    2019

  16. [24]

    and Bischof, H

    Donoser, M. and Bischof, H. Diffusion processes for retrieval revisited. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013

    2013

  17. [25]

    J., Cohen, E

    Evans, D. J., Cohen, E. G. D., and Morriss, G. P. Probability of second law violations in shearing steady states. Phys. Rev. Lett., 71: 0 2401--2404, Oct 1993

    1993

  18. [26]

    Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories

    Fei-Fei, L., Fergus, R., and Perona, P. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. In IEEE Conference on Computer Vision and Pattern Recognition Workshop, pp.\ 178--178. IEEE, 2004

    2004

  19. [27]

    Predict then propagate: Graph neural networks meet personalized pagerank

    Gasteiger, J., Bojchevski, A., and G \"u nnemann, S. Predict then propagate: Graph neural networks meet personalized pagerank. In International Conference on Learning Representations (ICLR), 2018

    2018

  20. [28]

    End-to-end learning of deep visual representations for image retrieval

    Gordo, A., Almazan, J., Revaud, J., and Larlus, D. End-to-end learning of deep visual representations for image retrieval. International Journal of Computer Vision, 124 0 (2): 0 237--254, 2017

    2017

  21. [29]

    Attention-based query expansion learning

    Gordo, A., Radenovic, F., and Berg, T. Attention-based query expansion learning. In European Conference on Computer Vision (ECCV), 2020

    2020

  22. [30]

    Pyretri: A pytorch-based library for unsupervised image retrieval by deep convolutional neural networks

    Hu, B., Song, R.-J., Wei, X.-S., Yao, Y., Hua, X.-S., and Liu, Y. Pyretri: A pytorch-based library for unsupervised image retrieval by deep convolutional neural networks. In Proceedings of ACM International Conference on Multimedia, pp.\ 4461--4464, 2020

    2020

  23. [31]

    Efficient diffusion on region manifolds: Recovering small objects with compact cnn representations

    Iscen, A., Tolias, G., Avrithis, Y., Furon, T., and Chum, O. Efficient diffusion on region manifolds: Recovering small objects with compact cnn representations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  24. [32]

    Fast spectral ranking for similarity search

    Iscen, A., Avrithis, Y., Tolias, G., Furon, T., and Chum, O. Fast spectral ranking for similarity search. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

    2018

  25. [33]

    Stochastic thermodynamic interpretation of information geometry

    Ito, S. Stochastic thermodynamic interpretation of information geometry. Phys. Rev. Lett., 121: 0 030605, Jul 2018

    2018

  26. [34]

    Scaling up visual and vision-language representation learning with noisy text supervision

    Jia, C., Yang, Y., Xia, Y., Chen, Y.-T., Parekh, Z., Pham, H., Le, Q., Sung, Y.-H., Li, Z., and Duerig, T. Scaling up visual and vision-language representation learning with noisy text supervision. In Proceedings of the 38th International Conference on Machine Learning. PMLR, 2021

    2021

  27. [35]

    A contextual dissimilarity measure for accurate and efficient image search

    Jégou, H., Harzallah, H., and Schmid, C. A contextual dissimilarity measure for accurate and efficient image search. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2007

    2007

  28. [36]

    Aggregating local image descriptors into compact codes

    Jégou, H., Perronnin, F., Douze, M., Sánchez, J., Pérez, P., and Schmid, C. Aggregating local image descriptors into compact codes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34 0 (9): 0 1704--1716, 2012

    2012

  29. [37]

    Self-taught metric learning without labels

    Kim, S., Kim, D., Cho, M., and Kwak, S. Self-taught metric learning without labels. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  30. [38]

    Correlation verification for image retrieval

    Lee, S., Seong, H., Lee, S., and Kim, E. Correlation verification for image retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  31. [39]

    Revisiting self-similarity: Structural embedding for image retrieval

    Lee, S., Lee, S., Seong, H., and Kim, E. Revisiting self-similarity: Structural embedding for image retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  32. [40]

    Li, J., Selvaraju, R., Gotmare, A., Joty, S., Xiong, C., and Hoi, S. C. H. Align before fuse: Vision and language representation learning with momentum distillation. In Advances in Neural Information Processing Systems, 2021

    2021

  33. [41]

    BLIP : Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Li, J., Li, D., Xiong, C., and Hoi, S. BLIP : Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Proceedings of the 39th International Conference on Machine Learning. PMLR, 2022

    2022

  34. [42]

    BLIP -2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Li, J., Li, D., Savarese, S., and Hoi, S. BLIP -2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In Proceedings of the 40th International Conference on Machine Learning. PMLR, 2023

    2023

  35. [43]

    Supervised metric learning to rank for retrieval via contextual similarity optimization

    Liao, C., Tsiligkaridis, T., and Kulis, B. Supervised metric learning to rank for retrieval via contextual similarity optimization. In Proceedings of the 40th International Conference on Machine Learning, Proceedings of Machine Learning Research. PMLR, 2023

    2023

  36. [44]

    Liu, C., Yu, G., Volkovs, M., Chang, C., Rai, H., Ma, J., and Gorti, S. K. Guided similarity separation for image retrieval. In Advances in Neural Information Processing Systems, 2019

    2019

  37. [45]

    Liu, H., Li, C., Wu, Q., and Lee, Y. J. Visual instruction tuning. In Advances in Neural Information Processing Systems, 2023

    2023

  38. [46]

    Lowe, D. G. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, 60: 0 91--110, 2004

    2004

  39. [47]

    Cluster-aware similarity diffusion for instance retrieval

    Luo, J., Yao, H., and Xu, C. Cluster-aware similarity diffusion for instance retrieval. In Proceedings of the 41st International Conference on Machine Learning. PMLR, 2024

    2024

  40. [48]

    Large-scale image retrieval with attentive deep local features

    Noh, H., Araujo, A., Sim, J., Weyand, T., and Han, B. Large-scale image retrieval with attentive deep local features. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017

    2017

  41. [49]

    Contextual similarity aggregation with self-attention for visual re-ranking

    Ouyang, J., Wu, H., Wang, M., Zhou, W., and Li, H. Contextual similarity aggregation with self-attention for visual re-ranking. In Advances in Neural Information Processing Systems, 2021

    2021

  42. [50]

    Object retrieval with large vocabularies and fast spatial matching

    Philbin, J., Chum, O., Isard, M., Sivic, J., and Zisserman, A. Object retrieval with large vocabularies and fast spatial matching. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2007

    2007

  43. [51]

    Lost in quantization: Improving particular object retrieval in large scale image databases

    Philbin, J., Chum, O., Isard, M., Sivic, J., and Zisserman, A. Lost in quantization: Improving particular object retrieval in large scale image databases. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008

    2008

  44. [52]

    R., and Tudisco, F

    Prokopchik, K., Benson, A. R., and Tudisco, F. Nonlinear feature diffusion on hypergraphs. In Proceedings of the International Conference on Machine Learning. PMLR, 2022

    2022

  45. [53]

    and Torralba, A

    Quattoni, A. and Torralba, A. Recognizing indoor scenes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 413--420. IEEE, 2009

    2009

  46. [54]

    Revisiting oxford and paris: Large-scale image retrieval benchmarking

    Radenovi \'c , F., Iscen, A., Tolias, G., Avrithis, Y., and Chum, O. Revisiting oxford and paris: Large-scale image retrieval benchmarking. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

    2018

  47. [55]

    Fine-tuning cnn image retrieval with no human annotation

    Radenovi \'c , F., Tolias, G., and Chum, O. Fine-tuning cnn image retrieval with no human annotation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41 0 (7): 0 1655--1668, 2019

    2019

  48. [56]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I. Learning transferable visual models from natural language supervision. In Proceedings of the International Conference on Machi...

    2021

  49. [57]

    S., Schumann, A., Eberle, A., and Stiefelhagen, R

    Sarfraz, M. S., Schumann, A., Eberle, A., and Stiefelhagen, R. A pose-sensitive embedding for person re-identification with expanded cross neighborhood re-ranking. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

    2018

  50. [58]

    Stochastic thermodynamics, fluctuation theorems and molecular machines

    Seifert, U. Stochastic thermodynamics, fluctuation theorems and molecular machines. Reports on Progress in Physics, 75 0 (12): 0 126001, nov 2012

    2012

  51. [59]

    Global features are all you need for image retrieval and reranking

    Shao, S., Chen, K., Karpur, A., Cui, Q., Araujo, A., and Cao, B. Global features are all you need for image retrieval and reranking. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2023

    2023

  52. [60]

    Object retrieval and localization with spatially-constrained similarity measure and k-nn re-ranking

    Shen, X., Lin, Z., Brandt, J., Avidan, S., and Wu, Y. Object retrieval and localization with spatially-constrained similarity measure and k-nn re-ranking. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012

    2012

  53. [61]

    X., Sbai, O., and Aubry, M

    Shen, X., Xiao, Y., Hu, S. X., Sbai, O., and Aubry, M. Re-ranking for image retrieval and transductive few-shot classification. In Advances in Neural Information Processing Systems, 2021

    2021

  54. [62]

    S., Tang, M., Sun, R., Yoon, J., Arik, S

    Su, H., Yen, H., Xia, M., Shi, W., Muennighoff, N., yu Wang, H., Haisu, L., Shi, Q., Siegel, Z. S., Tang, M., Sun, R., Yoon, J., Arik, S. O., Chen, D., and Yu, T. BRIGHT : A realistic and challenging benchmark for reasoning-intensive retrieval. In The Thirteenth International ...

    2025

  55. [63]

    BEIR : A heterogeneous benchmark for zero-shot evaluation of information retrieval models

    Thakur, N., Reimers, N., R \"u ckl \'e , A., Srivastava, A., and Gurevych, I. BEIR : A heterogeneous benchmark for zero-shot evaluation of information retrieval models. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2), 2021

    2021

  56. [64]

    Particular object retrieval with integral max-pooling of cnn activations

    Tolias, G., Sicre, R., and J \'e gou, H. Particular object retrieval with integral max-pooling of cnn activations. In International Conference on Learning Representations (ICLR), 2016

    2016

  57. [65]

    and Saito, K

    Van Vu, T. and Saito, K. Thermodynamic unification of optimal transport: Thermodynamic uncertainty relation, minimum dissipation, and thermodynamic speed limits. Phys. Rev. X, 13: 0 011013, Feb 2023

    2023

  58. [66]

    The caltech-ucsd birds-200-2011 dataset, 2011

    Wah, C., Branson, S., Welinder, P., Perona, P., and Belongie, S. The caltech-ucsd birds-200-2011 dataset, 2011

    2011

  59. [67]

    C-pack: Packaged resources to advance general chinese embedding, 2023

    Xiao, S., Liu, Z., Zhang, P., and Muennighoff, N. C-pack: Packaged resources to advance general chinese embedding, 2023

    2023

  60. [68]

    Efficient image retrieval via decoupling diffusion into online and offline processing

    Yang, F., Hinami, R., Matsui, Y., Ly, S., and Satoh, S. Efficient image retrieval via decoupling diffusion into online and offline processing. In Proceedings of the AAAI Conference on Artificial Intelligence, 2019

    2019

  61. [69]

    Dolg: Single-stage image retrieval with deep orthogonal fusion of local and global features

    Yang, M., He, D., Fan, M., Shi, B., Xue, X., Li, F., Ding, E., and Huang, J. Dolg: Single-stage image retrieval with deep orthogonal fusion of local and global features. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021

    2021

  62. [70]

    Yang, X., Koknar-Tezel, S., and Latecki, L. J. Locally constrained diffusion process on locally densified distance spaces with applications to shape retrieval. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009

    2009

  63. [71]

    Yang, X., Prasad, L., and Latecki, L. J. Affinity learning with diffusion on tensor product graph. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35 0 (1): 0 28--38, 2013

    2013

  64. [72]

    FILIP : Fine-grained interactive language-image pre-training

    Yao, L., Huang, R., Hou, L., Lu, G., Niu, M., Xu, H., Liang, X., Li, Z., Jiang, X., and Xu, C. FILIP : Fine-grained interactive language-image pre-training. In International Conference on Learning Representations, 2022

    2022

  65. [73]

    Contextually affinitive neighborhood refinery for deep clustering

    Yu, C., Shi, Y., and Wang, J. Contextually affinitive neighborhood refinery for deep clustering. In Advances in Neural Information Processing Systems, 2023

    2023

  66. [74]

    Zhang, S., Yang, M., Cour, T., Yu, K., and Metaxas, D. N. Query specific rank fusion for image retrieval. IEEE Transactions on Pattern Analysis and Machine Intelligence, 37 0 (4): 0 803--815, 2015

    2015

  67. [75]

    Understanding image retrieval re-ranking: A graph neural network perspective

    Zhang, X., Jiang, M., Zheng, Z., Tan, X., Ding, E., and Yang, Y. Understanding image retrieval re-ranking: A graph neural network perspective. arXiv preprint arXiv:2012.07620, 2020

    2012 arXiv

  68. [76]

    Graph convolution based efficient re-ranking for visual retrieval

    Zhang, Y., Qian, Q., Wang, H., Liu, C., Chen, W., and Wan, F. Graph convolution based efficient re-ranking for visual retrieval. IEEE Transactions on Multimedia, 2023

    2023

  69. [77]

    Re-ranking person re-identification with k-reciprocal encoding

    Zhong, Z., Zheng, L., Cao, D., and Li, S. Re-ranking person re-identification with k-reciprocal encoding. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  70. [78]

    Ranking on data manifolds

    Zhou, D., Weston, J., Gretton, A., Bousquet, O., and Sch\" o lkopf, B. Ranking on data manifolds. In Advances in Neural Information Processing Systems, 2003

    2003

  71. [79]

    Fusion with diffusion for robust visual tracking

    Zhou, Y., Bai, X., Liu, W., and Latecki, L. Fusion with diffusion for robust visual tracking. In Advances in Neural Information Processing Systems, 2012

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.