pith. machine review for the scientific record. sign in

arxiv: 2604.18037 · v1 · submitted 2026-04-20 · 💻 cs.CV

Recognition: unknown

HABIT: Chrono-Synergia Robust Progressive Learning Framework for Composed Image Retrieval

Zixu Li , Yupeng Hu , Zhiwei Chen , Shiqi Zhang , Qinlei Huang , Zhiheng Fu , Yinwei Wei
Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:47 UTC · model grok-4.3

classification 💻 cs.CV
keywords composed image retrievalnoisy tripletsrobust learningmutual informationprogressive learningnoise handlingimage retrieval
0
0 comments X

The pith

HABIT framework uses mutual information transition rates and dual model consistency to robustly retrieve images from noisy composed queries.

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

Composed image retrieval lets users find targets by combining a reference photo with text that describes desired changes, but real annotations often contain mismatches that degrade training. The paper introduces HABIT to solve the resulting noise problem by first measuring how cleanly each sample matches the intended change through the rate at which mutual information passes between the combined query and the target image. It then runs a progressive process in which a historical model and the current model cooperate, keeping reliable patterns while correcting unreliable ones. If this works, systems could maintain high accuracy even when a substantial share of the training triplets are noisy, making practical deployment in search and recommendation feasible without perfect labels.

Core claim

HABIT consists of the Mutual Knowledge Estimation Module, which quantifies sample cleanliness by calculating the Transition Rate of mutual information between the composed feature and the target image to identify samples aligned with intended modification semantics, and the Dual-consistency Progressive Learning Module, which introduces a collaborative mechanism between historical and current models to retain good habits and calibrate bad habits for robust adaptation under the presence of NTC.

What carries the argument

HABIT's Mutual Knowledge Estimation Module, which computes the transition rate of mutual information to score sample cleanliness, together with its Dual-consistency Progressive Learning Module, which simulates habit formation by letting historical and current models jointly retain reliable behaviors and correct unreliable ones.

Load-bearing premise

The transition rate of mutual information accurately estimates how well a sample aligns with the intended modification semantics, and the dual-consistency mechanism between historical and current models reliably enables adaptation to modification discrepancies.

What would settle it

An experiment that injects controlled noise known to break the correlation between mutual information transition rate and true semantic alignment, then checks whether HABIT's performance advantage over baselines disappears.

Figures

Figures reproduced from arXiv: 2604.18037 by Qinlei Huang, Shiqi Zhang, Yinwei Wei, Yupeng Hu, Zhiheng Fu, Zhiwei Chen, Zixu Li.

Figure 1
Figure 1. Figure 1: (a) presents an example of the CIR paradigm. (b) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: HABIT consists of two modules: (a) Mutual Knowledge Estimation and (b) Dual-consistency Progressive Learning. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: presents the top-5 results from HABIT and the SOTA robust CIR model TME on two CIR datasets. In the CIRR example ( [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comprehensive Performance Comparison Rank on CIRR and FashionIQ. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sensitivity to the hyperparameters (a) κ and (b) γ. Progressive Learning (DPL) module operates solely during training and introduces negligible inference overhead. Ad￾ditionally, HABIT increases training time by only approx￾imately 3.16% compared to w/o History (2.94s vs. 2.85s), while yielding notable performance gains on CIRR and FIQ. (5) From the architectural perspective, HABIT incor￾porates Mutual Kno… view at source ↗
Figure 6
Figure 6. Figure 6: Failure Cases on CIRR and FashionIQ datasets. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of the cleanliness estimation. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Similarity Matrix of TME and HABIT on CIRR [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
read the original abstract

Composed Image Retrieval (CIR) is a flexible image retrieval paradigm that enables users to accurately locate the target image through a multimodal query composed of a reference image and modification text. Although this task has demonstrated promising applications in personalized search and recommendation systems, it encounters a severe challenge in practical scenarios known as the Noise Triplet Correspondence (NTC) problem. This issue primarily arises from the high cost and subjectivity involved in annotating triplet data. To address this problem, we identify two central challenges: the precise estimation of composed semantic discrepancy and the insufficient progressive adaptation to modification discrepancy. To tackle these challenges, we propose a cHrono-synergiA roBust progressIve learning framework for composed image reTrieval (HABIT), which consists of two core modules. First, the Mutual Knowledge Estimation Module quantifies sample cleanliness by calculating the Transition Rate of mutual information between the composed feature and the target image, thereby effectively identifying clean samples that align with the intended modification semantics. Second, the Dual-consistency Progressive Learning Module introduces a collaborative mechanism between the historical and current models, simulating human habit formation to retain good habits and calibrate bad habits, ultimately enabling robust learning under the presence of NTC. Extensive experiments conducted on two standard CIR datasets demonstrate that HABIT significantly outperforms most methods under various noise ratios, exhibiting superior robustness and retrieval performance. Codes are available at https://github.com/Lee-zixu/HABIT

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes HABIT, a chrono-synergia robust progressive learning framework for Composed Image Retrieval (CIR) under the Noise Triplet Correspondence (NTC) problem. It consists of a Mutual Knowledge Estimation Module that quantifies sample cleanliness via the Transition Rate of mutual information between composed features and target images, and a Dual-consistency Progressive Learning Module that employs collaboration between historical and current models to simulate habit formation for retaining good adaptations and calibrating discrepancies. Experiments on two standard CIR datasets are reported to show that HABIT significantly outperforms most methods across various noise ratios, with claims of superior robustness and retrieval performance.

Significance. If the central empirical claims hold after addressing the validation gaps, the work would offer a practically relevant advance for CIR systems, where NTC arises frequently from costly and subjective triplet annotations. The progressive collaboration mechanism provides a conceptually distinct approach to robust learning, and the public code release aids reproducibility. The result would strengthen noise-robust multimodal retrieval if the mutual-information estimator is shown to isolate modification-aligned samples rather than incidental embedding properties.

major comments (2)
  1. [Mutual Knowledge Estimation Module] Mutual Knowledge Estimation Module: the claim that the Transition Rate of mutual information precisely estimates sample cleanliness aligned with intended modification semantics lacks supporting evidence such as correlation analysis against held-out clean labels or controls for confounding factors (e.g., image complexity or embedding norm). Without this, the separation of clean versus noisy triplets under NTC remains unverified.
  2. [Experiments] Experiments: the reported outperformance at high noise ratios is not accompanied by an ablation that isolates the contribution of the Mutual Knowledge Estimation Module from the Dual-consistency Progressive Learning Module. It is therefore unclear whether the gains derive from the mutual-information estimator or from the progressive collaboration alone.
minor comments (2)
  1. [Abstract] The abstract states the method outperforms 'most methods' but provides no quantitative metrics, dataset names, or baseline list; adding these would improve immediate readability.
  2. [Method] Notation for the Transition Rate and mutual-information quantities should be defined explicitly with equations in the method section to avoid ambiguity in the estimation procedure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript accordingly to strengthen the validation of our claims.

read point-by-point responses

  1. Referee: [Mutual Knowledge Estimation Module] Mutual Knowledge Estimation Module: the claim that the Transition Rate of mutual information precisely estimates sample cleanliness aligned with intended modification semantics lacks supporting evidence such as correlation analysis against held-out clean labels or controls for confounding factors (e.g., image complexity or embedding norm). Without this, the separation of clean versus noisy triplets under NTC remains unverified.

    Authors: We agree that direct supporting evidence for the Mutual Knowledge Estimation Module would strengthen the manuscript. While the end-to-end results under varying noise ratios provide indirect validation, we will add a dedicated analysis in the revision: correlation coefficients between the estimated transition rates and held-out clean labels on a controlled subset, along with controls for potential confounders such as image complexity (measured via entropy) and embedding norms. This will verify that the module isolates modification-aligned semantics. revision: yes

  2. Referee: [Experiments] Experiments: the reported outperformance at high noise ratios is not accompanied by an ablation that isolates the contribution of the Mutual Knowledge Estimation Module from the Dual-consistency Progressive Learning Module. It is therefore unclear whether the gains derive from the mutual-information estimator or from the progressive collaboration alone.

    Authors: We acknowledge this gap in the experimental design. The current results demonstrate the full framework's robustness, but to isolate contributions we will include new ablation studies in the revised manuscript. These will compare: (1) a baseline with only Dual-consistency Progressive Learning, (2) the Mutual Knowledge Estimation Module applied to a standard progressive learner, and (3) the full HABIT model. This will clarify the individual and synergistic effects, particularly at high noise ratios. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework is empirically defined and externally validated

full rationale

The paper defines two modules to address NTC: Mutual Knowledge Estimation via Transition Rate of mutual information between composed feature and target image, plus Dual-consistency Progressive Learning between historical and current models. These are introduced as novel components without any derivation chain, equations, or self-citations that reduce the claimed robustness or predictions back to fitted inputs or prior self-work by construction. Performance is shown via experiments on standard CIR datasets under varying noise ratios, providing independent empirical content. This matches the reader's assessment of no evident circular reasoning and qualifies as self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework depends on domain assumptions about the reliability of mutual information for cleanliness estimation and the effectiveness of historical-current model collaboration for habit-like learning, without independent evidence or formal derivation provided in the abstract.

axioms (2)
  • domain assumption The Transition Rate of mutual information between the composed feature and the target image accurately quantifies sample cleanliness for composed semantic discrepancy.
    This is the core mechanism of the Mutual Knowledge Estimation Module as stated in the abstract.
  • ad hoc to paper A collaborative mechanism between historical and current models can simulate human habit formation to retain good habits and calibrate bad habits for robust learning under NTC.
    This underpins the Dual-consistency Progressive Learning Module and is presented as a novel simulation without prior justification.

pith-pipeline@v0.9.0 · 5579 in / 1344 out tokens · 37719 ms · 2026-05-10T04:47:51.422662+00:00 · methodology

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

140 extracted references · 30 canonical work pages · 5 internal anchors

  1. [1]

    Wen, H.; Song, X.; Yin, J.; Wu, J.; Guan, W.; and Nie, L

  2. [2]

    Self-Training Boosted Multi-Factor Matching Net- work for Composed Image Retrieval.IEEE TPAMI

  3. [3]

    Xu, X.; Liu, Y .; Khan, S.; Khan, F.; Zuo, W.; Goh, R. S. M.; Feng, C.-M.; et al. 2024. Sentence-level Prompts Benefit Composed Image Retrieval. InICLR

    2024

  4. [4]

    Chen, Z.; Hu, Y .; Fu, Z.; Li, Z.; Huang, J.; Huang, Q.; and Wei, Y . 2026. INTENT: Invariance and Discrimination- aware Noise Mitigation for Robust Composed Image Re- trieval. InAAAI, volume 40, 20463–20471

    2026

  5. [5]

    Hu, Y .; Li, Z.; Chen, Z.; Huang, Q.; Fu, Z.; Xu, M.; and Nie, L. 2026. REFINE: Composed Video Retrieval via Shared and Differential Semantics Enhancement.ACM ToMM

    2026

  6. [6]

    Li, Z.; Hu, Y .; Chen, Z.; Huang, Q.; Qiu, G.; Fu, Z.; and Liu, M. 2026. ReTrack: Evidence-Driven Dual-Stream Di- rectional Anchor Calibration Network for Composed Video Retrieval. InAAAI, volume 40, 23373–23381

    2026

  7. [7]

    Liu, F.; Cheng, Z.; Zhu, L.; Gao, Z.; and Nie, L. 2021. Interest-aware message-passing GCN for recommendation. InACM WWW, 1296–1305

    2021

  8. [8]

    Yang, X.; Liu, D.; Zhang, H.; Luo, Y .; Wang, C.; and Zhang, J. 2024. Decomposing Semantic Shifts for Composed Image Retrieval. InAAAI, volume 38, 6576–6584

    2024

  9. [9]

    Jiang, X.; Wang, Y .; Li, M.; Wu, Y .; Hu, B.; and Qian, X

  10. [10]

    InACM SIGIR, 2177– 2187

    Cala: Complementary association learning for aug- menting comoposed image retrieval. InACM SIGIR, 2177– 2187

  11. [11]

    Cheng, Z.; Lai, L.; Liu, Y .; Cheng, K.; and Qi, X. 2026. En- hancing Financial Report Question-Answering: A Retrieval- Augmented Generation System with Reranking Analysis. arXiv preprint arXiv:2603.16877

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  12. [12]

    Pu, R.; Qin, Y .; Song, X.; Peng, D.; Ren, Z.; and Sun, Y

  13. [13]

    SHE: Streaming-media Hashing Retrieval. InICML

  14. [14]

    Xie, Z. 2026. CONQUER: Context-Aware Representation with Query Enhancement for Text-Based Person Search. arXiv preprint arXiv:2601.18625

    work page arXiv 2026

  15. [15]

    Xie, Z.; Liu, X.; Zhang, B.; Lin, Y .; Cai, S.; and Jin, T. 2026. HVD: Human Vision-Driven Video Represen- tation Learning for Text-Video Retrieval.arXiv preprint arXiv:2601.16155

    work page arXiv 2026

  16. [16]

    Chen, K.; Fang, P.; and Xue, H. 2025. DePro: Domain Ensemble using Decoupled Prompts for Universal Cross- Domain Retrieval. InSIGIR, SIGIR ’25, 958–967

    2025

  17. [17]

    Chen, K.; Fang, P.; and Xue, H. 2025. Multi-Modal Inter- active Agent Layer for Few-Shot Universal Cross-Domain Retrieval and Beyond. InNeurIPS

    2025

  18. [18]

    Wang, Y .; Fu, T.; Xu, Y .; Ma, Z.; Xu, H.; Du, B.; Lu, Y .; Gao, H.; Wu, J.; and Chen, J. 2024. TWIN-GPT: digital twins for clinical trials via large language model.ACM ToMM

    2024

  19. [19]

    Chen, W.; Wu, L.; Hu, Y .; Li, Z.; Cheng, Z.; Qian, Y .; Zhu, L.; Hu, Z.; Liang, L.; Tang, Q.; et al. 2025. AutoNeural: Co-Designing Vision-Language Models for NPU Inference. arXiv preprint arXiv:2512.02924

    work page arXiv 2025

  20. [20]

    Jia, S.; Zhu, N.; Zhong, J.; Zhou, J.; Zhang, H.; Hwang, J.- N.; and Li, L. 2026. RAM: Recover Any 3D Human Motion in-the-Wild. arXiv:2603.19929

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    Li, L.; Jia, S.; Wang, J.; Jiang, Z.; Zhou, F.; Dai, J.; Zhang, T.; Wu, Z.; and Hwang, J.-N. 2025. Human Motion Instruc- tion Tuning. InCVPR

    2025

  22. [22]

    Li, L.; Jia, S.; and Hwang, J.-N. 2026. Multiple Human Motion Understanding. InAAAI, volume 40, 6297–6305

    2026

  23. [23]

    Ni, C.; Wang, X.; Zhu, Z.; Wang, W.; Li, H.; Zhao, G.; Li, J.; Qin, W.; Huang, G.; and Mei, W. 2025. Wonderturbo: Gen- erating interactive 3d world in 0.72 seconds.arXiv preprint arXiv:2504.02261

    work page arXiv 2025

  24. [24]

    Qiu, X.; Wu, X.; Lin, Y .; Guo, C.; Hu, J.; and Yang, B. 2025. DUET: Dual Clustering Enhanced Multivariate Time Series Forecasting. InSIGKDD, 1185–1196

    2025

  25. [25]

    Zhao, Z. 2024. Balf: Simple and efficient blur aware local feature detector. InWACV, 3362–3372

    2024

  26. [26]

    Zhang, F.; Gu, Z.; and Wang, H. 2026. Decoding with struc- tured awareness: integrating directional, frequency-spatial, and structural attention for medical image segmentation. In AAAI, volume 40, 12421–12429

    2026

  27. [27]

    Liu, J.; Zhuo, D.; Feng, Z.; Zhu, S.; Peng, C.; Liu, Z.; and Wang, H. 2024. Dvlo: Deep visual-lidar odometry with local-to-global feature fusion and bi-directional structure alignment. InECCV, 475–493. Springer

    2024

  28. [28]

    Jiang, G.; Zhang, T.; Li, D.; Zhao, Z.; Li, H.; Li, M.; and Wang, H. 2025. STG-Avatar: Animatable Human Avatars via Spacetime Gaussian.arXiv preprint arXiv:2510.22140

    work page arXiv 2025

  29. [29]

    Yuan, H.; Li, X.; Dai, J.; You, X.; Sun, Y .; and Ren, Z. 2025. Deep Streaming View Clustering. InICML

    2025

  30. [30]

    Lu, S.; Liu, Y .; and Kong, A. W.-K. 2023. Tf-icon: Diffusion-based training-free cross-domain image compo- sition. InICCV, 2294–2305

    2023

  31. [31]

    Zhou, S.; Cao, Y .; Nie, J.; Fu, Y .; Zhao, Z.; Lu, X.; and Wang, S. 2026. Comptrack: Information bottleneck-guided low-rank dynamic token compression for point cloud track- ing. InAAAI, volume 40, 13773–13781

    2026

  32. [32]

    Zhou, Z.; Lu, S.; Leng, S.; Zhang, S.; Lian, Z.; Yu, X.; and Kong, A. W.-K. 2025. DragFlow: Unleashing DiT Priors with Region Based Supervision for Drag Editing.arXiv preprint arXiv:2510.02253

    work page arXiv 2025

  33. [33]

    Lan, Y .; Xu, S.; Su, C.; Ye, R.; Peng, D.; and Sun, Y . 2025. Multi-view Hashing Classification. InACM MM, 2122– 2130

    2025

  34. [34]

    Yu, Z.; IDRIS, M. Y . I.; Wang, P.; and Qureshi, R. 2025. CoTextor: Training-Free Modular Multilingual Text Editing via Layered Disentanglement and Depth-Aware Fusion. In NeurIPS

    2025

  35. [35]

    Liao, B.; Zhao, Z.; Li, H.; Zhou, Y .; Zeng, Y .; Li, H.; and Liu, P. 2025. Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World. InCVPR, 15823–15832

    2025

  36. [36]

    S.; Sheng, Z.; and Yang, B

    Qiu, X.; Hu, J.; Zhou, L.; Wu, X.; Du, J.; Zhang, B.; Guo, C.; Zhou, A.; Jensen, C. S.; Sheng, Z.; and Yang, B. 2024. TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods. InVLDB, 2363–2377

    2024

  37. [37]

    Li, L.; Lu, S.; Ren, Y .; and Kong, A. W.-K. 2025. Set you straight: Auto-steering denoising trajectories to sidestep un- wanted concepts.arXiv preprint arXiv:2504.12782

    work page arXiv 2025

  38. [38]

    Duan, S.; Wu, W.; Hu, P.; Ren, Z.; Peng, D.; and Sun, Y

  39. [39]

    CoPINN: Cognitive physics-informed neural net- works. InICML

  40. [40]

    Zhou, S.; Nie, J.; Zhao, Z.; Cao, Y .; and Lu, X. 2025. Focus- track: One-stage focus-and-suppress framework for 3d point cloud object tracking. InACM MM, 7366–7375

    2025

  41. [41]

    Xie, Z.; Wang, C.; Wang, Y .; Cai, S.; Wang, S.; and Jin, T

  42. [42]

    InEMNLP, 5259–5270

    Chat-driven text generation and interaction for person retrieval. InEMNLP, 5259–5270

  43. [43]

    Liu, P. 2024. Unsupervised corrupt data detection for text training.ESWA, 248: 123335

    2024

  44. [44]

    Xie, Z.; Zhang, B.; Lin, Y .; and Jin, T. 2026. Delving deeper: Hierarchical visual perception for robust video-text retrieval.arXiv preprint arXiv:2601.12768

    work page arXiv 2026

  45. [45]

    Gu, R.; Jia, S.; Ma, Y .; Zhong, J.; Hwang, J.-N.; and Li, L

  46. [46]

    InACM MM, 9026–9034

    MoCount: Motion-Based Repetitive Action Counting. InACM MM, 9026–9034

  47. [47]

    Li, L.; Jia, S.; Wang, J.; An, Z.; Li, J.; Hwang, J.-N.; and Be- longie, S. 2025. Chatmotion: A multimodal multi-agent for human motion analysis.arXiv preprint arXiv:2502.18180

    work page arXiv 2025

  48. [48]

    Jia, S.; and Li, L. 2024. Adaptive masking enhances visual grounding.arXiv preprint arXiv:2410.03161

    work page arXiv 2024

  49. [49]

    Liu, L.; Chen, S.; Jia, S.; Shi, J.; Jiang, Z.; Jin, C.; Zongkai, W.; Hwang, J.-N.; and Li, L. 2024. Graph canvas for control- lable 3d scene generation.arXiv preprint arXiv:2412.00091

    work page arXiv 2024

  50. [50]

    Liu, P.; Yang, J.; Wang, L.; Wang, S.; Hao, Y .; and Bai, H

  51. [51]

    InCIKM, 4099–4104

    Retrieval-Based Unsupervised Noisy Label Detection on Text Data. InCIKM, 4099–4104

  52. [52]

    Yang, Q.; Chen, Z.; Hu, Y .; Li, Z.; Fu, Z.; and Nie, L. 2026. STABLE: Efficient Hybrid Nearest Neighbor Search via Magnitude-Uniformity and Cardinality-Robustness.arXiv preprint arXiv:2604.01617

    work page arXiv 2026

  53. [53]

    Yang, Q.; Lv, P.; Li, Y .; Zhang, S.; Chen, Y .; Chen, Z.; Li, Z.; and Hu, Y . 2026. ERASE: Bypassing Collaborative Detec- tion of AI Counterfeit Via Comprehensive Artifacts Elimi- nation.IEEE TDSC, 1–18

    2026

  54. [54]

    T.; Peng, X.; and Hu, P

    Li, S.; He, C.; Liu, X.; Zhou, J. T.; Peng, X.; and Hu, P. 2025. Learning with Noisy Triplet Correspondence for Composed Image Retrieval. InCVPR, 19628–19637

    2025

  55. [55]

    Liu, M.; Wang, X.; Nie, L.; He, X.; Chen, B.; and Chua, T.-S. 2018. Attentive moment retrieval in videos. InACM SIGIR, 15–24

    2018

  56. [56]

    Hu, Y .; Liu, M.; Su, X.; Gao, Z.; and Nie, L. 2021. Video moment localization via deep cross-modal hashing.IEEE TIP, 30: 4667–4677

    2021

  57. [57]

    Liu, M.; Wang, X.; Nie, L.; Tian, Q.; Chen, B.; and Chua, T.-S. 2018. Cross-modal moment localization in videos. In ACM MM, 843–851

    2018

  58. [58]

    Hu, Y .; Wang, K.; Liu, M.; Tang, H.; and Nie, L. 2023. Se- mantic collaborative learning for cross-modal moment lo- calization.ACM TOIS, 42(2): 1–26

    2023

  59. [59]

    Liu, F.; Liu, Y .; Chen, H.; Cheng, Z.; Nie, L.; and Kankan- halli, M. 2025. Understanding Before Recommendation: Se- mantic Aspect-Aware Review Exploitation via Large Lan- guage Models.ACM TOIS, 43(2)

    2025

  60. [60]

    Liu, P.; Wang, X.; Cui, Z.; and Ye, W. 2025. Queries Are Not Alone: Clustering Text Embeddings for Video Search. InSIGIR, 874–883

    2025

  61. [61]

    Wen, H.; Zhang, X.; Song, X.; Wei, Y .; and Nie, L. 2023. Target-guided composed image retrieval. InACM MM, 915– 923

    2023

  62. [62]

    Li, Z.; Chen, Z.; Wen, H.; Fu, Z.; Hu, Y .; and Guan, W

  63. [63]

    ENCODER: Entity Mining and Modification Relation Binding for Composed Image Retrieval. InAAAI

  64. [64]

    Chen, Z.; Hu, Y .; Li, Z.; Fu, Z.; Wen, H.; and Guan, W

  65. [65]

    InACM MM, 6143–6152

    HUD: Hierarchical Uncertainty-Aware Disambigua- tion Network for Composed Video Retrieval. InACM MM, 6143–6152. ACM

  66. [66]

    Chen, Z.; Hu, Y .; Li, Z.; Fu, Z.; Song, X.; and Nie, L

  67. [67]

    InACM MM, 6113–6122

    OFFSET: Segmentation-based Focus Shift Revision for Composed Image Retrieval. InACM MM, 6113–6122. ACM

  68. [68]

    Mu, C.; Yang, E.; and Deng, C. 2025. Meta-Guided Adap- tive Weight Learner for Noisy Correspondence. InACM SIGIR, 968–978

    2025

  69. [69]

    Zha, Q.; Liu, X.; Cheung, Y .-m.; Peng, S.-J.; Xu, X.; and Wang, N. 2025. UCPM: Uncertainty-Guided Cross-Modal Retrieval with Partially Mismatched Pairs.IEEE TIP

    2025

  70. [70]

    Wu, H.; Gao, Y .; Guo, X.; Al-Halah, Z.; Rennie, S.; Grau- man, K.; and Feris, R. 2021. Fashion iq: A new dataset to- wards retrieving images by natural language feedback. In CVPR, 11307–11317

    2021

  71. [71]

    Guo, X.; Wu, H.; Cheng, Y .; Rennie, S.; Tesauro, G.; and Feris, R. S. 2018. Dialog-based Interactive Image Retrieval. InNeurIPS, 676–686. MIT Press

    2018

  72. [72]

    P.; Yu, J.; and Yang, W

    Song, Z.; Luo, R.; Ma, L.; Tang, Y .; Chen, Y .-P. P.; Yu, J.; and Yang, W. 2025. Temporal Coherent Object Flow for Multi-Object Tracking. InAAAI, volume 39, 6978–6986

    2025

  73. [73]

    Zhou, S.; Yuan, Z.; Yang, D.; Hu, X.; Qian, J.; and Zhao, Z

  74. [74]

    InCVPR, 27336–27345

    Pillarhist: A quantization-aware pillar feature encoder based on height-aware histogram. InCVPR, 27336–27345

  75. [75]

    Liu, J.; Wang, G.; Ye, W.; Jiang, C.; Han, J.; Liu, Z.; Zhang, G.; Du, D.; and Wang, H. 2024. DifFlow3D: Toward Ro- bust Uncertainty-Aware Scene Flow Estimation with Itera- tive Diffusion-Based Refinement. InCVPR, 15109–15119

    2024

  76. [76]

    Liu, J.; Ye, W.; Wang, G.; Jiang, C.; Pan, L.; Han, J.; Liu, Z.; Zhang, G.; and Wang, H. 2025. DifFlow3D: Hierarchi- cal Diffusion Models for Uncertainty-Aware 3D Scene Flow Estimation.IEEE TPAMI

    2025

  77. [77]

    Jiang, K.; Dong, H.; Kang, Z.; Zhu, Z.; and Song, G. 2026. FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models. arXiv:2604.02967

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  78. [78]

    Gao, D.; Lu, S.; Walters, S.; Zhou, W.; Chu, J.; Zhang, J.; Zhang, B.; Jia, M.; Zhao, J.; Fan, Z.; et al. 2024. EraseAny- thing: Enabling Concept Erasure in Rectified Flow Trans- formers.arXiv preprint arXiv:2412.20413

    work page arXiv 2024

  79. [79]

    P.; and Yang, W

    Song, Z.; Luo, R.; Yu, J.; Chen, Y .-P. P.; and Yang, W. 2023. Compact transformer tracker with correlative masked mod- eling. InAAAI, volume 37, 2321–2329

    2023

  80. [80]

    Yu, Z.; Idris, M. Y . I.; and Wang, P. 2025. Visualizing Our Changing Earth: A Creative AI Framework for Democratiz- ing Environmental Storytelling Through Satellite Imagery. InNeurIPS

    2025

Showing first 80 references.