Pith. sign in

REVIEW 4 major objections 5 minor 64 references

Multiple low-resolution faces of the same person can be fused into one identity prior that drives progressive super-resolution and better re-identification.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 16:55 UTC pith:MHYO6F2S

load-bearing objection Clean multi-shot face SR pipeline that actually works; novelty is the joint cascade+attention design, not a new principle. the 4 major comments →

arxiv 2607.28130 v1 pith:MHYO6F2S submitted 2026-07-30 cs.CV

Collaborative Feature Aggregation for Face Super-Resolution and Robust Re-Identification

Juheon Hwang , Taewan Kim , Jiwoo Kang This is my paper
classification cs.CV
keywords face super-resolutioncollaborative identity aggregationtransformer attentioncascade residual restorationperson re-identificationmulti-view facesvideo surveillance
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Single-image face super-resolution often blurs or invents details when the input is severely degraded, because one frame simply lacks enough information. This paper shows that a set of sequential or multi-view faces of the same person can be turned into a single, shared identity embedding by a transformer attention module; that embedding is then fed, at every stage, into a cascade residual network that gradually rebuilds high-resolution detail. The same unified embedding also serves as a robust descriptor for person re-identification under low resolution. Experiments on video and multi-view face datasets, plus the MARS re-identification benchmark, report consistent gains over recent single-image, video, and multi-view baselines. The practical claim is that joint identity reconstruction and progressive restoration from multiple observations measurably improves both visual quality and downstream matching.

Core claim

When several low-resolution faces of one identity are available, a transformer-based collaborative aggregation can produce a unified identity feature that, when injected into a multi-stage residual cascade, recovers sharper, more identity-consistent high-resolution faces than methods that rely on a single image or on non-identity multi-frame fusion, and the same feature improves video person re-identification under degradation.

What carries the argument

Transformer-based collaborative identity aggregation (multi-head self-attention over ArcFace-style embeddings of N faces) that yields a unified identity prior f_ID, which is re-estimated and injected at each stage of a residual cascade restoration model that predicts and adds image deltas.

Load-bearing premise

That several degraded views of the same person always contain a common facial identity signal strong enough for attention-based fusion to recover a useful prior rather than noise or conflicting cues.

What would settle it

Run the full pipeline on paired low-resolution sequences where ground-truth identity is known but faces are heavily occluded or extreme-pose; if the collaborative model’s PSNR, identity cosine similarity, and re-ID mAP fall to or below the single-image cascade baseline, the shared-identity premise fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Surveillance systems can treat short multi-view or multi-frame face tracks as a single collaborative SR-and-re-ID unit instead of independent frames.
  • Identity-consistent face restoration becomes feasible at 8× upscaling from 32×32 inputs without external landmark or codebook priors.
  • The same cascade-plus-unified-embedding architecture transfers, with a generic backbone, to full-body pedestrian re-identification under low resolution.
  • Progressive residual stages reduce the risk of identity drift that appears when generative priors are substituted for the true subject.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the shared-identity signal is the true bottleneck, then adaptive weighting that down-weights occluded or extreme-pose frames inside the attention module should further lift both SR fidelity and re-ID rank-1.
  • The same collaborative prior could be attached to any residual or diffusion SR backbone, suggesting the aggregation step is more portable than the particular cascade decoder.
  • In single-image regimes the method collapses to its weaker “Ours-” variant, so a practical system would need a cheap detector that decides when a second observation is worth waiting for.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a collaborative face super-resolution framework that aggregates identity features from N sequential or multi-view low-resolution faces via a transformer multi-head self-attention module (Eqs. 1, 4), then injects the unified identity prior into a residual cascade restoration model (CRM; Eqs. 2–5, M=3) that progressively predicts deltas toward the HR target. Image L1 and ArcFace identity L1 losses supervise end-to-end training. The same identity pathway is reused for video person re-identification on MARS by attaching FC layers and swapping ArcFace for a ResNet backbone. Experiments on VFHQ, CelebV-HQ (sequential), Multiface (multi-view), and degraded MARS faces report gains over IDM, VQFR, MVSR, SAVSR and several re-ID baselines (Tables 1–5), with ablations on cascade, unified ID, N, and identity backbones.

Significance. If the claimed gains hold under realistic multi-observation surveillance conditions, the work is a useful systems contribution: it couples multi-image identity unification with progressive residual SR and shows transfer to re-ID. Strengths include clear ablations (no unified ID, no cascade, N=1..5, ArcFace/AdaFace/TopoFR), standard external metrics (PSNR/SSIM/LPIPS/ID/FID, mAP/rank-1), public datasets, and qualitative error maps. The contribution is incremental rather than foundational—attention over frozen face embeddings plus cascade residuals is a natural combination—but the joint SR+re-ID framing and multi-view/sequential evaluation are of practical interest to the surveillance CV community.

major comments (4)
  1. [§3.1, Eqs. (1),(4); §5] §3.1–3.3, Eqs. (1) and (4): The central claim rests on multi-head attention over ArcFace (or similar) features yielding a reliably stronger unified prior f_ID / f^k_ID than any single ϕ(I_i), including under 8× degradation, pose change, and MARS-level corruption. Tables 1 and 4 show large drops without unified ID, but the paper never reports a direct quality measure of the unified embedding (e.g., cosine similarity of f_ID to GT identity, or agreement across views) nor a controlled corruption/pose ablation that isolates when unification helps versus collapses. §5 acknowledges extreme pose/occlusion as limitations; without such measurements the load-bearing assumption remains untested and the SR/re-ID gains cannot be attributed specifically to successful identity unification.
  2. [§4.2, §4.3.4] §4.2 and §4.3.1–4.3.4: All primary quantitative SR results use synthetic bicubic 32→256 (8×) downsampling. Real surveillance degradation (compression, blur, noise, demosaicing) differs substantially; the only real-degraded evidence is qualitative MARS face crops (Fig. 6). This weakens the claim of robustness for surveillance. At minimum, add quantitative results under standard blind/real degradation pipelines (or on native low-quality faces with paired or no-reference metrics) so that gains versus IDM/VQFR/MVSR/SAVSR are not confined to bicubic conditions.
  3. [Table 3, §4.3.5] Table 3 (§4.3.5): Re-ID gains on MARS are marginal versus the strongest baseline (mAP 83.9 vs BAU 83.6; rank-1 90.3 vs 89.1). The protocol mixes 2× SR with re-ID fine-tuning, replaces the face network with ImageNet-ResNet, and reports single-run point estimates with no error bars or multiple seeds. The tiny mAP delta does not by itself establish that collaborative identity unification transfers beyond face SR; either strengthen the protocol (full-body multi-camera protocols, statistical significance, ablation of SR vs identity-aggregation contributions) or temper the joint SR+re-ID claim in the abstract and conclusions.
  4. [§4.3.1, Tables 1–2] §4.3.1–4.3.3, Tables 1–2: Comparison fairness needs clarification. MVSR and SAVSR are multi-view/video SR methods not designed with face-identity conditioning; VQFR/IDM are single-image. It is unclear whether multi-frame baselines received the same N=2 paired inputs, identical train/test splits, and face-cropped preprocessing, or were run off-the-shelf. Without a protocol paragraph specifying input cardinality, fine-tuning, and face alignment for every baseline, the SOTA claim is hard to audit. Please document the exact multi-image protocol used for each competitor.
minor comments (5)
  1. [Throughout] Typos and wording: “supperresolution” (Fig. 2 caption, §5, Data availability), “Futhermore”, “manneer”, “ide ntity”, “frech´ et”. Sweep the manuscript.
  2. [Fig. 2, §3] Fig. 2 notation mixes I, bold I, and script variants; f_ID vs f^k_ID vs ϕ is dense. A short symbol table would help.
  3. [§3.4] λ_ID is set to 1.0 with no sensitivity study; a one-row ablation would strengthen §3.4.
  4. [§3.2, §5] M=3 and N=2 are empirical (§3.2, Fig. 9) but computational cost of the cascade and attention is not reported (FLOPs/latency vs single-pass baselines).
  5. [§2] Related work on video face SR and multi-frame fusion could more clearly position against landmark/segmentation priors already cited [11–14] and recent identity-conditioned diffusion SR.

Circularity Check

0 steps flagged

No circularity: empirical multi-frame SR/re-ID architecture evaluated on external benchmarks against held-out GT; nothing reduces by construction to its inputs.

full rationale

This is a standard computer-vision methods paper. The claimed gains come from (i) multi-head self-attention over ArcFace (or alternative) identity embeddings from N inputs (Eqs. 1, 4), (ii) residual cascade restoration conditioned on the unified prior (Eqs. 2–5), and (iii) end-to-end training with L1 image + identity losses against ground-truth images and features (Eqs. 6–9). Performance is reported with ordinary external metrics (PSNR/SSIM/LPIPS/ID/FID, mAP/rank-1) on public test splits (VFHQ, CelebV-HQ, Multiface, MARS) versus independent baselines. Ablations (Table 4, Fig. 8–9) and backbone swaps (Table 5) are empirical sensitivity checks, not definitional identities. Design constants (M=3, N=2, λ_ID=1) are stated hyperparameters, not fitted quantities re-labeled as predictions. There is no self-definitional loop, no fitted-input-called-prediction, no load-bearing self-citation uniqueness theorem, and no renaming of a known closed-form result. The load-bearing scientific assumption (recoverable shared identity under degradation) is an empirical hypothesis tested by the experiments, not a circular reduction. Score 0 is therefore appropriate.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 3 invented entities

The claim rests on standard deep-learning practice plus a few domain assumptions about multi-shot identity and several hand-chosen training constants. No new physical entities; invented pieces are architectural modules whose only evidence is the paper’s own ablations and benchmarks.

free parameters (5)
  • cascade iterations M = 3
    Number of residual CRM stages; set empirically to 3 without a principled schedule (§3.2).
  • number of collaborative inputs N = 2
    Default pair of images chosen after a 1–5 sweep showing diminishing returns (Fig. 9); balances cost vs gain.
  • identity loss weight λ_ID = 1.0
    Balances L1 image loss and L1 identity feature loss; fixed to 1.0 without sensitivity study (§3.4).
  • Adam learning rate and schedule length = 1e-4, 4000 epochs, bs=4
    Training hyperparameters fixed at 1e-4 for 4000 epochs, batch size 4 (§4.2).
  • SR scale / LR synthesis = 8× bicubic
    Defines the task as bicubic 32×32→256×256 (8×); central quantitative tables depend on this degradation model (§4.2).
axioms (5)
  • domain assumption Pretrained face-recognition embeddings (ArcFace by default) are sufficiently identity-discriminative on low-resolution inputs to serve as SR conditioning and re-ID features.
    Invoked in §3.1 and Eq. 1; backbone swap in Table 5 assumes the same role for AdaFace/TopoFR.
  • domain assumption Multiple temporal or multi-view observations of one person contain a shared identity signal that multi-head self-attention can aggregate without primary/secondary image designation.
    Core premise of the collaborative identity network (§3.1, Eq. 1) and of re-estimating f^k_ID each stage (Eq. 4).
  • domain assumption Progressive residual refinement (I^{k+1}=I^k+ΔI^k) with skip-connected encoder and MLP implicit decoder stably recovers high-frequency face detail from upsampled LR inputs.
    Cascade restoration model design in §3.2 and Eq. 2; ablation without cascade in Table 4/Fig. 8.
  • domain assumption Bicubic downsampling of curated high-quality face videos is an adequate proxy for evaluating surveillance face SR, with MARS crops as a limited real-degradation check.
    Implementation and comparison protocol §4.2–4.3; real degraded qualitative study only in §4.3.4.
  • standard math Standard multi-head self-attention and UNet/MLP building blocks behave as in the cited transformer and INR literature.
    Uses Vaswani et al. attention and LIIF-style MLP decoding without modification proofs (§3).
invented entities (3)
  • Unified facial identity embedding f_ID (and per-stage f^k_ID) no independent evidence
    purpose: Single order-invariant identity prior aggregated from N face embeddings to condition every cascade stage and support re-ID.
    Defined in Eqs. 1 and 4; central object the method optimizes for. Evidence is internal metrics/ablations only.
  • Cascade restoration model (CRM) with identity-conditioned residual stages no independent evidence
    purpose: Iteratively predict RGB residuals toward HR faces while receiving the unified identity prior.
    Architectural invention in §3.2–3.3; not a separately validated module outside this paper’s experiments.
  • Transformer-based collaborative identity network no independent evidence
    purpose: Model inter-image correlations among face ID features without designating a reference view.
    Fig. 2(a) and §3.1; novelty claim hinges on this module plus cascade coupling.

pith-pipeline@v1.2.0-daily-grok45 · 23054 in / 3953 out tokens · 80711 ms · 2026-07-31T16:55:01.325192+00:00 · methodology

0 comments
read the original abstract

We propose a novel collaborative approach for face super-resolution (SR) and robust person re-identification from sequential or multi-view facial images. Traditional SR methods often suffer from blurring and distortion in faces recovered from poor-quality images due to low resolution. Image- and video-based facial SR methods using facial landmarks or segmentation also have similar challenges. To overcome these limitations, we leverage multiple correlated facial observations, across time or viewpoints, by introducing a transformer-based collaborative feature aggregation method that unifies identity features from multi-sequence or multi-view data. This allows faces in multiple sequences of an individual to contribute to accurately estimating common facial features. Furthermore, we propose a cascade SR network to progressively restore the high-resolution image of the target's face with gradual facial feature unification. The unified identity representation is further utilized in person re-identification scenarios, enabling accurate matching even under severe image degradation. The exhaustive experimental results and comparisons show that our method outperforms other state-of-the-art methods, demonstrating consistent improvements in both face super-resolution and re-identification performance. Our work highlights the effectiveness of joint identity reconstruction and progressive image restoration from multiple facial inputs in enhancing downstream visual recognition tasks.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

64 extracted references · 2 linked inside Pith

  1. [1]

    Multimedia Systems 30(6), 377 (2024)

    Wu, H., Hao, X., Yang, L.: Mini-infrared imaging system image super-resolution via symmetric channel change and deep residual network. Multimedia Systems 30(6), 377 (2024)

    2024

  2. [2]

    Multimedia Systems 27, 821–836 (2021)

    Liu, J., Ge, J., Xue, Y., He, W., Sun, Q., Li, S.: Multi-scale skip-connection network for image super-resolution. Multimedia Systems 27, 821–836 (2021)

    2021

  3. [3]

    Springer (2021)

    Bellandi, V., Chehri, A., Cuomo, S., Jeon, G.: Special issue on deep learning for emerging big multimedia super-resolution. Springer (2021)

    2021

  4. [4]

    Multimedia Systems 30(5), 250 (2024)

    Wang, D., Wu, K.: Anomaly detection in surveillance videos using transformer with margin learning. Multimedia Systems 30(5), 250 (2024)

    2024

  5. [5]

    Multimedia systems 29(1), 59–71 (2023)

    Liu, W., Cao, J., Zhu, Y., Liu, B., Zhu, X.: Real-time anomaly detection on surveillance video with two-stream spatio-temporal generative model. Multimedia systems 29(1), 59–71 (2023)

    2023

  6. [6]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Liang, J., Cao, J., Sun, G., Zhang, K., Van Gool, L., Timofte, R.: SwinIR: Image restoration using swin transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1833–1844 (2021). IEEE

    2021

  7. [7]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Wang, X., Xie, L., Dong, C., Shan, Y.: Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1905–1914 (2021). IEEE

    1905

  8. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Chen, Y., Liu, S., Wang, X.: Learning continuous image representation with local implicit image function. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8628–8638 (2021). IEEE

    2021

  9. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Gao, S., Liu, X., Zeng, B., Xu, S., Li, Y., Luo, X., Liu, J., Zhen, X., Zhang, B.: Implicit diffusion models for continuous super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10021–10030 (2023). IEEE

    2023

  10. [10]

    In: Proceedings of European Conference on Computer Vision, pp

    Zheng, L., Bie, Z., Sun, Y., Wang, J., Su, C., Wang, S., Tian, Q.: MARS: A video benchmark for large-scale person re-identification. In: Proceedings of European Conference on Computer Vision, pp. 868–884 (2016). Springer 23

    2016

  11. [11]

    IEEE Transactions on Image Processing 29, 2150–2165 (2019)

    Grm, K., Scheirer, W.J., ˇStruc, V.: Face hallucination using cascaded super- resolution and identity priors. IEEE Transactions on Image Processing 29, 2150–2165 (2019)

    2019

  12. [12]

    In: Proceedings of the IEEE International Joint Conference on Biometrics, pp

    Zhang, C., Liu, Z.: Face super-resolution with progressive embedding of multi- scale face priors. In: Proceedings of the IEEE International Joint Conference on Biometrics, pp. 1–8 (2022). IEEE

    2022

  13. [13]

    Frontiers in Neurorobotics 16, 5 (2022)

    Teng, Z., Yu, X., Wu, C.: Blind face restoration via multi-prior collaboration and adaptive feature fusion. Frontiers in Neurorobotics 16, 5 (2022)

    2022

  14. [14]

    In: Proceedings of the European Conference on Computer Vision, pp

    Gu, Y., Wang, X., Xie, L., Dong, C., Li, G., Shan, Y., Cheng, M.-M.: VQFR: Blind face restoration with vector-quantized dictionary and parallel decoder. In: Proceedings of the European Conference on Computer Vision, pp. 126–143 (2022). Springer

    2022

  15. [15]

    In: Proceedings of the Asian Conference on Computer Vision, pp

    Bian, S., Li, H., Yu, F., Liu, J., Changjun, S., Tang, Y.: F APN: Face alignment propagation network for face video super-resolution. In: Proceedings of the Asian Conference on Computer Vision, pp. 1–16 (2022)

    2022

  16. [16]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Yu, F., Li, H., Bian, S., Tang, Y.: An efficient network design for face video super-resolution. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1513–1520 (2021). IEEE

    2021

  17. [17]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Xin, J., Wang, N., Li, J., Gao, X., Li, Z.: Video face super-resolution with motion- adaptive feedback cell. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 12468–12475 (2020)

    2020

  18. [18]

    Neural Computing and Applications 33, 13505–13523 (2021)

    Farooq, M., Dailey, M.N., Mahmood, A., Moonrinta, J., Ekpanyapong, M.: Human face super-resolution on poor quality surveillance video footage. Neural Computing and Applications 33, 13505–13523 (2021)

    2021

  19. [19]

    In: Pro- ceedings of the European Conference on Computer Vision Workshops, pp

    Wang, X., Yu, K., Wu, S., Gu, J., Liu, Y., Dong, C., Qiao, Y., Change Loy, C.: ESRGAN: Enhanced super-resolution generative adversarial networks. In: Pro- ceedings of the European Conference on Computer Vision Workshops, pp. 63–79 (2019). Springer

    2019

  20. [20]

    In: Proceedings of the Advances in Neural Information Processing Systems, vol

    Yariv, L., Gu, J., Kasten, Y., Lipman, Y.: Volume rendering of neural implicit sur- faces. In: Proceedings of the Advances in Neural Information Processing Systems, vol. 34, pp. 4805–4815 (2021)

    2021

  21. [21]

    In: Proceedings of the Advances in Neural Information Processing Systems, vol

    Wang, P., Liu, L., Liu, Y., Theobalt, C., Komura, T., Wang, W.: NeuS: Learning neural implicit surfaces by volume rendering for multi-view Reconstruction. In: Proceedings of the Advances in Neural Information Processing Systems, vol. 34, pp. 27171–27183 (2021)

    2021

  22. [22]

    Communications of the ACM 65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., 24 Ng, R.: NeRF: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM 65(1), 99–106 (2021)

    2021

  23. [23]

    In: Pro- ceedings of the Advances in Neural Information Processing Systems, vol

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. In: Pro- ceedings of the Advances in Neural Information Processing Systems, vol. 33, pp. 6840–6851 (2020)

    2020

  24. [24]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Chen, X., Liu, X., Liu, W., Zhang, X.-P., Zhang, Y., Mei, T.: Explainable per- son re-identification with attribute-guided metric distillation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 11813–11822 (2021)

    2021

  25. [25]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Choi, S., Kim, T., Jeong, M., Park, H., Kim, C.: Meta batch-instance normaliza- tion for generalizable person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 3425–3435 (2021)

    2021

  26. [26]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Huang, Y., Wu, Q., Xu, J., Zhong, Y., Zhang, Z.: Clothing status aware- ness for long-term person re-identification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 11895–11904 (2021)

    2021

  27. [27]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Li, H., Wu, G., Zheng, W.-S.: Combined depth space based architecture search for person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6729–6738 (2021)

    2021

  28. [28]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Park, H., Lee, S., Lee, J., Ham, B.: Learning by aligning: Visible-infrared per- son re-identification using cross-modal correspondences. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 12046–12055 (2021)

    2021

  29. [29]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Ren, M., He, L., Liao, X., Liu, W., Wang, Y., Tan, T.: Learning instance- level spatial-temporal patterns for person re-identification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 14930–14939 (2021)

    2021

  30. [30]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Wang, X., Li, S., Liu, M., Wang, Y., Roy-Chowdhury, A.K.: Multi-expert adver- sarial attack detection in person re-identification using context inconsistency. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 15097–15107 (2021)

    2021

  31. [31]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Zheng, Y., Tang, S., Teng, G., Ge, Y., Liu, K., Qin, J., Qi, D., Chen, D.: Online pseudo label generation by hierarchical cluster dynamics for adaptive person re- identification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 8371–8381 (2021)

    2021

  32. [32]

    In: Proceedings of the European Conference on Computer Vision, pp

    Gu, X., Chang, H., Ma, B., Zhang, H., Chen, X.: Appearance-preserving 3D con- volution for video-based person re-identification. In: Proceedings of the European Conference on Computer Vision, pp. 228–243 (2020). Springer 25

    2020

  33. [33]

    In: Proceedings of the European Conference on Computer Vision, pp

    Hou, R., Chang, H., Ma, B., Shan, S., Chen, X.: Temporal complementary learn- ing for video person re-identification. In: Proceedings of the European Conference on Computer Vision, pp. 388–405 (2020). Springer

    2020

  34. [34]

    In: Proceedings of the European Conference on Computer Vision, pp

    Shim, M., Ho, H.-I., Kim, J., Wee, D.: Read: Reciprocal attention discriminator for image-to-video re-identification. In: Proceedings of the European Conference on Computer Vision, pp. 335–350 (2020). Springer

    2020

  35. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Yan, Y., Qin, J., Chen, J., Liu, L., Zhu, F., Tai, Y., Shao, L.: Learning multi- granular hypergraphs for video-based person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2899–2908 (2020)

    2020

  36. [36]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhang, Z., Lan, C., Zeng, W., Chen, Z.: Multi-granularity reference-aided atten- tive feature aggregation for video-based person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10407–10416 (2020)

    2020

  37. [37]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Eom, C., Lee, G., Lee, J., Ham, B.: Video-based person re-identification with spatial and temporal memory networks. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 12036–12045 (2021)

    2021

  38. [38]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Wang, Y., Zhang, P., Gao, S., Geng, X., Lu, H., Wang, D.: Pyramid spatial- temporal aggregation for video-based person re-identification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 12026–12035 (2021)

    2021

  39. [39]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Aich, A., Zheng, M., Karanam, S., Chen, T., Roy-Chowdhury, A.K., Wu, Z.: Spatio-temporal representation factorization for video-based person re- identification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 152–162 (2021)

    2021

  40. [40]

    In: Proceedings of the Advances in Neural Information Processing Systems, vol

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. In: Proceedings of the Advances in Neural Information Processing Systems, vol. 30 (2017)

    2017

  41. [41]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Deng, J., Guo, J., Xue, N., Zafeiriou, S.: ArcFace: Additive angular margin loss for deep face recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4690–4699 (2019). IEEE

    2019

  42. [42]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Karras, T., Laine, S., Aila, T.: A style-based generator architecture for generative adversarial networks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4401–4410 (2019). IEEE

    2019

  43. [43]

    In: Proceedings of the International Conference on Learning Representations (2018)

    Karras, T., Aila, T., Laine, S., Lehtinen, J.: Progressive growing of GANs for improved quality, stability, and variation. In: Proceedings of the International Conference on Learning Representations (2018). https://openreview.net/forum? 26 id=Hk99zCeAb

    2018

  44. [44]

    In: Proceedings of the INTERSPEECH 2017, pp

    Nagrani, A., Chung, J.S., Zisserman, A.: VoxCeleb: A large-scale speaker iden- tification dataset. In: Proceedings of the INTERSPEECH 2017, pp. 2616–2620 (2017)

    2017

  45. [45]

    In: Proceedings of the INTERSPEECH 2018, pp

    Chung, J., Nagrani, A., Zisserman, A.: VoxCeleb2: Deep speaker recognition. In: Proceedings of the INTERSPEECH 2018, pp. 1086–1090 (2018)

    2018

  46. [46]

    Computer Speech & Language 60, 101027 (2020)

    Nagrani, A., Chung, J.S., Xie, W., Zisserman, A.: Voxceleb: Large-scale speaker verification in the wild. Computer Speech & Language 60, 101027 (2020)

    2020

  47. [47]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Xie, L., Wang, X., Zhang, H., Dong, C., Shan, Y.: VFHQ: A high-quality dataset and benchmark for video face super-resolution. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 657–666 (2022). IEEE

    2022

  48. [48]

    In: Proceedings of the European Conference on Computer Vision, pp

    Zhu, H., Wu, W., Zhu, W., Jiang, L., Tang, S., Zhang, L., Liu, Z., Loy, C.C.: CelebV-HQ: A large-scale video facial attributes dataset. In: Proceedings of the European Conference on Computer Vision, pp. 650–667 (2022). Springer

    2022

  49. [49]

    arXiv preprint arXiv:2207.11243 (2022)

    Wuu, C.-h., Zheng, N., Ardisson, S., Bali, R., Belko, D., Brockmeyer, E., Evans, L., Godisart, T., Ha, H., Hypes, A., et al.: Multiface: A dataset for neural face rendering. arXiv preprint arXiv:2207.11243 (2022)

    Pith/arXiv arXiv 2022

  50. [50]

    arXiv preprint arXiv:1412.6980 (2014)

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

    Pith/arXiv arXiv 2014

  51. [51]

    In: Proceedings of the Advances in Neural Information Processing Systems Workshop (2017)

    Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic differentiation in PyTorch. In: Proceedings of the Advances in Neural Information Processing Systems Workshop (2017)

    2017

  52. [52]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Kim, M., Jain, A.K., Liu, X.: AdaFace: Quality adaptive margin for face recog- nition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18750–18759 (2022)

    2022

  53. [53]

    In: Proceedings of the Advances in Neural Information Processing Systems

    Dan, J., Liu, Y., Deng, J., Xie, H., Li, S., Sun, B., Luo, S.: TopoFR: A closer look at topology alignment on face recognition. In: Proceedings of the Advances in Neural Information Processing Systems

  54. [54]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Li, Z., Liu, H., Shang, F., Liu, Y., Wan, L., Feng, W.: SA VSR: Arbitrary-scale video super-resolution via a learned scale-adaptive network. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 3288–3296 (2024)

    2024

  55. [55]

    In: Proceedings of the 30th ACM 27 International Conference on Multimedia, pp

    Cheng, R., Sun, Y., Yan, B., Tan, W., Ma, C.: Geometry-aware reference syn- thesis for multi-view image super-resolution. In: Proceedings of the 30th ACM 27 International Conference on Multimedia, pp. 6083–6093 (2022)

    2022

  56. [56]

    In: Proceedings of the Advances in Neural Information Processing Systems, vol

    Van Den Oord, A., Vinyals, O., et al.: Neural discrete representation learning. In: Proceedings of the Advances in Neural Information Processing Systems, vol. 30, pp. 6309–6318 (2017)

    2017

  57. [57]

    IEEE Transactions on Image Processing 13(4), 600–612 (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assess- ment: From error visibility to structural similarity. IEEE Transactions on Image Processing 13(4), 600–612 (2004)

    2004

  58. [58]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 586–595 (2018). IEEE

    2018

  59. [59]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Chan, E.R., Lin, C.Z., Chan, M.A., Nagano, K., Pan, B., De Mello, S., Gallo, O., Guibas, L.J., Tremblay, J., Khamis, S., et al.: Efficient geometry-aware 3D generative adversarial networks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 16123–16133 (2022). IEEE

    2022

  60. [60]

    In: Proceedings of the Advances in Neural Information Processing Systems, vol

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.: GANs trained by a two time-scale update rule converge to a local nash equilibrium. In: Proceedings of the Advances in Neural Information Processing Systems, vol. 30, pp. 6629–6640 (2017)

    2017

  61. [61]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recogni- tion. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778 (2016)

    2016

  62. [62]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Bai, S., Ma, B., Chang, H., Huang, R., Chen, X.: Salient-to-broad transition for video person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7339–7348 (2022)

    2022

  63. [63]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhu, H., Budhwant, P., Zheng, Z., Nevatia, R.: SEAS: ShapE-aligned supervision for person re-identification. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 164–174 (2024)

    2024

  64. [64]

    In: Proceedings of the Advances in Neural Information Processing Sys- tems (2024)

    Cho, Y., Kim, J., Kim, W.J., Jung, J., Yoon, S.-E.: Generalizable person re-identification via balancing alignment and uniformity. In: Proceedings of the Advances in Neural Information Processing Sys- tems (2024). https://proceedings.neurips.cc/paper files/paper/2024/hash/ 53fba4404ebecf9730dc8919b71d4d22-Abstract-Conference.html 28

    2024