The reviewed record of science sign in
Pith

arxiv: 2606.09360 · v1 · pith:U44PKPFH · submitted 2026-06-08 · cs.CV

ExDet: Open-Domain Open-Vocabulary Detection with Cross-modal Extrapolation and Rectification

Yupeng Zhang , Yuzhong Feng , Ruize Han , Zhiwei Chen , Wei Feng , Liang Wan This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-27 16:51 UTCgrok-4.3pith:U44PKPFHrecord.jsonopen to challenge →

classification cs.CV
keywords open-domain open-vocabulary detectioncross-modal extrapolationdetector-compatible rectificationvision-language modelsdomain generalizationobject detection
0
0 comments X

The pith

ExDet lets existing detectors handle novel categories and unseen domains by generating text-based visual prototypes and rectifying outputs at inference without retraining.

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

The paper introduces ExDet as a lightweight framework that adds cross-category and cross-domain generalization to existing open-vocabulary detectors. It avoids retraining full models from scratch by leveraging vision-language models to create proxy visual prototypes from text descriptions. These prototypes train a rectification module that adjusts detector representations toward the original training distribution, while an updated region proposal step improves object recall. The approach claims state-of-the-art results on OD-LVIS, OV-LVIS, Objects365, and MSOSB benchmarks at low added cost.

Core claim

ExDet is a category-domain collaborative generalization framework for open-domain open-vocabulary detection consisting of Text-Guided Extrapolation (TGE) that exploits the DeltaSpace property of vision-language models to infer category- and domain-aware proxy visual prototypes from text, a Detector-Compatible Rectification (DCR) module learned from the TGE-generated prototypes in a detector training-free and real-data-free manner and inserted after the classification head at inference to rectify representations toward a detector-compatible source-domain visual distribution, and ExRPN that recalibrates proposal scores by combining semantic similarity with RPN confidence, thereby enhancing cla

What carries the argument

The ExDet framework built around TGE for text-to-visual prototype extrapolation, DCR for training-free representation rectification toward source-domain statistics, and ExRPN for semantic-RPN proposal recalibration.

If this is right

  • Existing detectors gain improved classification accuracy for novel categories and domain-shifted objects.
  • ExRPN raises recall for novel and domain-shifted objects while supplying better inputs to classification and DCR.
  • The full pipeline reaches state-of-the-art on OD-LVIS, OV-LVIS, Objects365, and MSOSB.
  • The DCR component adds no detector retraining or real-data requirements, keeping added cost low.

Where Pith is reading between the lines

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

  • The same prototype-generation and rectification pattern could be tested on related tasks such as open-vocabulary segmentation or instance segmentation.
  • Because DCR requires no real images, the method might support rapid adaptation in settings where collecting target-domain data is expensive or restricted.
  • The reliance on DeltaSpace properties of vision-language models suggests checking whether similar extrapolation works when swapping in other pretrained multimodal models.

Load-bearing premise

The Detector-Compatible Rectification module, derived only from text-generated prototypes, can successfully shift representations of novel categories and unseen domains to match the source-domain visual distribution the detector expects.

What would settle it

A controlled test in which adding the DCR module after the classification head produces no accuracy gain or a drop on held-out novel-category and domain-shifted detection tasks compared with the unmodified base detector.

Figures

Figures reproduced from arXiv: 2606.09360 by Liang Wan, Ruize Han, Wei Feng, Yupeng Zhang, Yuzhong Feng, Zhiwei Chen.

Figure 1
Figure 1. Figure 1: PCA visualization of visual extrapolation for novel [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of ExDet. Built upon a frozen F-ViT detector, ExDet is a lightweight category–domain generalization [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of OV-LVIS and OD-LVIS object embeddings. After applying DCR, the OD-LVIS embeddings (yellow) are [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of our method, DVtor, and GT under diverse domain shifts and challenging imaging conditions. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Open-domain open-vocabulary detection (ODOVD) requires detectors to generalize to both novel categories and unseen domains, making it more challenging than open-vocabulary detection. Existing methods typically train open-vocabulary detectors together with domain generalization modules from scratch, leading to high training cost. we propose ExDet, a lightweight category-domain collaborative generalization framework for ODOVD that enhances the cross-category and cross-domain generalization of existing detectors. ExDet consists of Text-Guided Extrapolation (TGE), a lightweight Detector-Compatible Rectification (DCR) module, and ExRPN. Specifically, TGE exploits the DeltaSpace property of vision-language models (VLMs) to infer category- and domain-aware proxy visual prototypes from text. DCR is learned from the TGE-generated prototypes in a detector training-free and real-data-free manner, and is inserted after the classification head at inference to rectify representations toward a detector-compatible source-domain visual distribution, thereby enhancing classification for targets from novel categories and unseen domains. ExRPN recalibrates proposal scores by combining semantic similarity with RPN confidence, improving recall for novel and domain-shifted objects while providing better support for subsequent classification and DCR. ExDet achieves SOTA performance on OD-LVIS, OV-LVIS, Objects365, and MSOSB.

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

3 major / 2 minor

Summary. The paper proposes ExDet, a lightweight category-domain collaborative generalization framework for open-domain open-vocabulary detection (ODOVD). It consists of Text-Guided Extrapolation (TGE) that uses the DeltaSpace property of VLMs to infer category- and domain-aware proxy visual prototypes from text, a Detector-Compatible Rectification (DCR) module learned solely from these TGE prototypes in a detector training-free and real-data-free manner and inserted after the classification head at inference to align novel-category/unseen-domain features to the source visual distribution, and ExRPN that recalibrates proposal scores via semantic similarity and RPN confidence. The framework is claimed to enhance cross-category and cross-domain generalization of existing detectors and to achieve SOTA performance on OD-LVIS, OV-LVIS, Objects365, and MSOSB.

Significance. If the central claims hold, the work would be significant for reducing training costs in ODOVD by enabling plug-in generalization modules that operate without real data or detector retraining. The training-free, real-data-free nature of DCR, if validated, would represent a practical advance over methods that jointly train detectors and domain generalization components from scratch.

major comments (3)
  1. [Abstract] Abstract: the SOTA claims on OD-LVIS, OV-LVIS, Objects365, and MSOSB are asserted without any reported baselines, ablations, quantitative metrics, or error analysis, rendering it impossible to assess whether the data support the generalization improvements attributed to DCR and ExRPN.
  2. [Method (DCR module)] DCR description: the claim that DCR, trained exclusively on TGE-generated text-derived prototypes, successfully rectifies real detector outputs for novel categories and unseen domains rests on the unverified assumption that these synthetic prototypes span the necessary source-domain visual statistics (texture, lighting, etc.); this is load-bearing for both the SOTA numbers and the 'lightweight/training-free' framing and requires concrete validation such as feature distribution comparisons or controlled ablations.
  3. [Method (TGE and DCR)] TGE and DCR interaction: no derivation or analysis is supplied showing that the DeltaSpace extrapolation produces prototypes whose statistics are sufficient to learn a rectification mapping that generalizes beyond the synthetic manifold to actual detector feature distributions.
minor comments (2)
  1. [Method (TGE)] Clarify the precise mathematical definition of the DeltaSpace property and the extrapolation procedure used to generate prototypes.
  2. [Abstract] The abstract states that ExRPN 'provides better support for subsequent classification and DCR' but does not specify how the recalibrated scores interact with the rectification step.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the SOTA claims on OD-LVIS, OV-LVIS, Objects365, and MSOSB are asserted without any reported baselines, ablations, quantitative metrics, or error analysis, rendering it impossible to assess whether the data support the generalization improvements attributed to DCR and ExRPN.

    Authors: The detailed baselines, ablations, quantitative metrics, and error analysis appear in Sections 4 and 5. The abstract is a concise summary and does not contain these numbers. We will revise the abstract to include key mAP improvements and SOTA margins on the cited benchmarks. revision: yes

  2. Referee: [Method (DCR module)] DCR description: the claim that DCR, trained exclusively on TGE-generated text-derived prototypes, successfully rectifies real detector outputs for novel categories and unseen domains rests on the unverified assumption that these synthetic prototypes span the necessary source-domain visual statistics (texture, lighting, etc.); this is load-bearing for both the SOTA numbers and the 'lightweight/training-free' framing and requires concrete validation such as feature distribution comparisons or controlled ablations.

    Authors: End-to-end gains on real benchmarks support the practical utility of DCR, yet direct statistical validation (e.g., feature distribution overlap) between TGE prototypes and real source features is absent. We will add t-SNE visualizations and quantitative distribution metrics comparing TGE prototypes to real detector features in the revised version. revision: yes

  3. Referee: [Method (TGE and DCR)] TGE and DCR interaction: no derivation or analysis is supplied showing that the DeltaSpace extrapolation produces prototypes whose statistics are sufficient to learn a rectification mapping that generalizes beyond the synthetic manifold to actual detector feature distributions.

    Authors: Section 3.1 describes the DeltaSpace property and Section 3.2 explains DCR training on the resulting prototypes. A formal derivation of statistical sufficiency for cross-manifold generalization is not provided. We will add a short analysis subsection deriving the conditions under which TGE prototypes enable the observed rectification on real features. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The provided manuscript text presents ExDet as an empirical framework (TGE exploiting DeltaSpace of VLMs, DCR learned from generated prototypes in a detector-free manner, ExRPN for proposal recalibration) that achieves reported SOTA on listed benchmarks. No equations, derivations, or parameter-fitting steps are shown that reduce any claimed prediction or result to its own inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems. The central claims rest on the design and empirical outcomes rather than any self-referential reduction, making the work self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified assumption that VLMs possess a usable DeltaSpace property for prototype generation and that the proposed rectification works without real data or retraining.

axioms (1)
  • domain assumption Vision-language models possess a DeltaSpace property that permits inference of category- and domain-aware proxy visual prototypes from text.
    Explicitly invoked by TGE in the abstract.

pith-pipeline@v0.9.1-grok · 5781 in / 1198 out tokens · 22181 ms · 2026-06-27T16:51:55.725348+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

64 extracted references · 14 canonical work pages · 1 internal anchor

  1. [1]

    Hanoona Bangalath, Muhammad Maaz, Muhammad Uzair Khattak, Salman H Khan, and Fahad Shahbaz Khan. 2022. Bridging the gap between object and image-level representations for open-vocabulary detection.Advances in Neural Information Processing Systems35 (2022), 33781–33794

    2022

  2. [2]

    Haibo Chen, Zhoujie Wang, Lei Zhao, Jun Li, and Jian Yang. 2025. Trtst: Arbitrary high-quality text-guided style transfer with transformers.IEEE Transactions on Image Processing(2025)

    2025

  3. [3]

    Sheng Cheng, Tejas Gokhale, and Yezhou Yang. 2023. Adversarial Bayesian Augmentation for Single-Source Domain Generalization. InProceedings of the IEEE/CVF International Conference on Computer Vision. 11400–11410

    2023

  4. [4]

    Tianheng Cheng, Lin Song, Yixiao Ge, Wenyu Liu, Xinggang Wang, and Ying Shan

  5. [5]

    InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yolo-world: Real-time open-vocabulary object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 16901– 16911

  6. [6]

    Muhammad Sohail Danish, Muhammad Haris Khan, Muhammad Akhtar Munir, M Saquib Sarfraz, and Mohsen Ali. 2024. Improving single domain-generalized object detection: A focus on diversification and alignment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 17732–17742

    2024

  7. [7]

    Yu Du, Fangyun Wei, Zihe Zhang, Miaojing Shi, Yue Gao, and Guoqi Li. 2022. Learning to prompt for open-vocabulary object detection with vision-language model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 14084–14093

    2022

  8. [8]

    Qi Fan, Mattia Segu, Yu-Wing Tai, Fisher Yu, Chi-Keung Tang, Bernt Schiele, and Dengxin Dai. 2023. Towards robust object detection invariant to real-world do- main shifts. InThe Eleventh International Conference on Learning Representations (ICLR 2023)

    2023

  9. [9]

    Tejas Gokhale, Rushil Anirudh, Jayaraman J Thiagarajan, Bhavya Kailkhura, Chitta Baral, and Yezhou Yang. 2023. Improving diversity with adversarially learned transformations for domain generalization. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 434–443

    2023

  10. [10]

    Xiuye Gu, Tsung-Yi Lin, Weicheng Kuo, and Yin Cui. 2021. Open-vocabulary object detection via vision and language knowledge distillation.arXiv preprint arXiv:2104.13921(2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  11. [11]

    Agrim Gupta, Piotr Dollar, and Ross Girshick. 2019. Lvis: A dataset for large vocabulary instance segmentation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 5356–5364

    2019

  12. [12]

    Jianhong Han, Yupei Wang, and Liang Chen. 2025. Style-Adaptive Detection Transformer for Single-Source Domain Generalized Object Detection.arXiv preprint arXiv:2504.20498(2025)

    work page arXiv 2025

  13. [13]

    Seongmin Hwang, Daeyoung Han, and Moongu Jeon. 2025. DG-DETR: Toward Domain Generalized Detection Transformer.arXiv preprint arXiv:2504.19574 (2025)

    work page arXiv 2025

  14. [14]

    Joonhyun Jeong, Geondo Park, Jayeon Yoo, Hyungsik Jung, and Heesu Kim

  15. [15]

    InProceedings of the AAAI Conference on Artificial Intelligence, Vol

    ProxyDet: Synthesizing Proxy Novel Classes via Classwise Mixup for Open-Vocabulary Object Detection. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 2462–2470

  16. [16]

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. 2021. Scaling up visual and vision- language representation learning with noisy text supervision. InInternational conference on machine learning. 4904–4916

    2021

  17. [17]

    Aishwarya Kamath, Mannat Singh, Yann LeCun, Gabriel Synnaeve, Ishan Misra, and Nicolas Carion. 2021. Mdetr-modulated detection for end-to-end multi- modal understanding. InProceedings of the IEEE/CVF international conference on computer vision. 1780–1790

    2021

  18. [18]

    Dahun Kim, Anelia Angelova, and Weicheng Kuo. 2023. Region-aware pretraining for open-vocabulary object detection with vision transformers. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 11144–11154

    2023

  19. [19]

    Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson, Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalantidis, Li-Jia Li, David A Shamma, et al

  20. [20]

    Visual genome: Connecting language and vision using crowdsourced dense image annotations.International journal of computer vision123 (2017), 32–73

    2017

  21. [21]

    Weicheng Kuo, Yin Cui, Xiuye Gu, AJ Piergiovanni, and Anelia Angelova. 2022. F- vlm: Open-vocabulary object detection upon frozen vision and language models. arXiv preprint arXiv:2209.15639(2022)

    work page arXiv 2022

  22. [22]

    Wooju Lee, Dasol Hong, Hyungtae Lim, and Hyun Myung. 2024. Object-aware domain generalization for object detection. Inproceedings of the AAAI conference on artificial intelligence, Vol. 38. 2947–2955

    2024

  23. [23]

    Jiaming Li, Jiacheng Zhang, Jichang Li, Ge Li, Si Liu, Liang Lin, and Guanbin Li. 2024. Learning background prompts to discover implicit knowledge for open vocabulary object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 16678–16687

    2024

  24. [24]

    Liangqi Li, Jiaxu Miao, Dahu Shi, Wenming Tan, Ye Ren, Yi Yang, and Shiliang Pu

  25. [25]

    InProceedings of the IEEE/CVF International Conference on Computer Vision

    Distilling detr with visual-linguistic knowledge for open-vocabulary object detection. InProceedings of the IEEE/CVF International Conference on Computer Vision. 6501–6510

  26. [26]

    Chuang Lin, Zehuan Yuan, Sicheng Zhao, Peize Sun, Changhu Wang, and Jianfei Cai. 2021. Domain-invariant disentangled network for generalizable object detection. InProceedings of the IEEE/CVF international conference on computer vision. 8771–8780

    2021

  27. [27]

    Xudong Liu, Zikun Chen, Ruowei Jiang, Ziyi Wu, Kejia Yin, Han Zhao, Parham Aarabi, and Igor Gilitschenski. 2025. S 2Edit: Text-Guided Image Editing with Precise Semantic and Spatial Control.arXiv preprint arXiv:2507.04584(2025)

    work page arXiv 2025

  28. [28]

    Yajing Liu, Shijun Zhou, Xiyao Liu, Chunhui Hao, Baojie Fan, and Jiandong Tian. 2024. Unbiased faster r-cnn for single-source domain generalized object detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 28838–28847

    2024

  29. [29]

    Yueming Lyu, Tianwei Lin, Fu Li, Dongliang He, Jing Dong, and Tieniu Tan. 2023. Deltaedit: Exploring text-free training for text-driven image manipulation.arXiv preprint arXiv:2303.06285(2023)

    work page arXiv 2023

  30. [30]

    Yueming Lyu, Kang Zhao, Bo Peng, Yue Jiang, Yingya Zhang, and Jing Dong

  31. [31]

    DeltaSpace: A Semantic-aligned Feature Space for Flexible Text-guided Image Editing.arXiv preprint arXiv:2310.08785(2023)

    work page arXiv 2023

  32. [32]

    M Minderer, A Gritsenko, A Stone, M Neumann, D Weissenborn, A Dosovit- skiy, A Mahendran, A Arnab, M Dehghani, Z Shen, et al . 2022. Simple open- vocabulary object detection with vision transformers. arxiv 2022.arXiv preprint arXiv:2205.062302 (2022)

    work page arXiv 2022

  33. [33]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. InInternational conference on machine learning. 8748–8763

    2021

  34. [34]

    Zhijie Rao, Jingcai Guo, Luyao Tang, Yue Huang, Xinghao Ding, and Song Guo

  35. [35]

    Srcd: Semantic reasoning with compound domains for single-domain generalized object detection.IEEE Transactions on Neural Networks and Learning Systems(2024)

    2024

  36. [36]

    Tal Ridnik, Emanuel Ben-Baruch, Asaf Noy, and Lihi Zelnik-Manor. 2021. Imagenet-21k pretraining for the masses.arXiv preprint arXiv:2104.10972(2021)

    work page arXiv 2021

  37. [37]

    Shuai Shao, Zeming Li, Tianyuan Zhang, Chao Peng, Gang Yu, Xiangyu Zhang, Jing Li, and Jian Sun. 2019. Objects365: A large-scale, high-quality dataset for ob- ject detection. InProceedings of the IEEE/CVF international conference on computer vision. 8430–8439

    2019

  38. [38]

    Piyush Sharma, Nan Ding, Sebastian Goodman, and Radu Soricut. 2018. Con- ceptual captions: A cleaned, hypernymed, image alt-text dataset for automatic image captioning. InProceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2556–2565

    2018

  39. [39]

    Vidit Vidit, Martin Engilberge, and Mathieu Salzmann. 2023. Clip the gap: A single domain generalization approach for object detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 3219–3229

    2023

  40. [40]

    Ao Wang, Lihao Liu, Hui Chen, Zijia Lin, Jungong Han, and Guiguang Ding. 2025. Yoloe: Real-time seeing anything.arXiv preprint arXiv:2503.07465(2025)

    work page arXiv 2025

  41. [41]

    Junjie Wang, Bin Chen, Bin Kang, Yulin Li, Weizhi Xian, Yichi Chen, and Yong Xu. 2025. Ov-dquo: Open-vocabulary detr with denoising text query training and open-world unknown objects supervision. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 7762–7770

    2025

  42. [42]

    Junjie Wang, Bin Chen, Yulin Li, Bin Kang, Yichi Chen, and Zhuotao Tian. 2025. DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception. InProceed- ings of the Computer Vision and Pattern Recognition Conference. 14824–14834

    2025

  43. [43]

    Jiacheng Wang, Ping Liu, Jingen Liu, and Wei Xu. 2023. Text-guided eyeglasses manipulation with spatial constraints.IEEE Transactions on Multimedia26 (2023), 4375–4388

    2023

  44. [44]

    Luting Wang, Yi Liu, Penghui Du, Zihan Ding, Yue Liao, Qiaosong Qi, Biaolong Chen, and Si Liu. 2023. Object-aware distillation pyramid for open-vocabulary object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 11186–11196

    2023

  45. [45]

    Aming Wu and Cheng Deng. 2022. Single-domain generalized object detection in urban scene via cyclic-disentangled self-distillation. InProceedings of the IEEE/CVF Conference on computer vision and pattern recognition. 847–856

    2022

  46. [46]

    Fan Wu, Jinling Gao, Lanqing Hong, Xinbing Wang, Chenghu Zhou, and Nanyang Ye. 2024. G-NAS: Generalizable Neural Architecture Search for Single Domain Generalization Object Detection. InProceedings of the AAAI Conference on Artifi- cial Intelligence, Vol. 38. 5958–5966

    2024

  47. [47]

    Fei Wu, Yongheng Ma, Hao Jin, Xiao-Yuan Jing, and Guo-Ping Jiang. 2023. MFE- CLIP: CLIP with mapping-fusion embedding for text-guided image editing.IEEE Signal Processing Letters31 (2023), 116–120

    2023

  48. [48]

    Size Wu, Wenwei Zhang, Lumin Xu, Sheng Jin, Xiangtai Li, Wentao Liu, and Chen Change Loy. 2023. Clipself: Vision transformer distills itself for open- vocabulary dense prediction.arXiv preprint arXiv:2310.01403(2023)

    work page arXiv 2023

  49. [49]

    Xiaoshi Wu, Feng Zhu, Rui Zhao, and Hongsheng Li. 2023. Cora: Adapting clip for open-vocabulary detection with region prompting and anchor pre-matching. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7031–7040

    2023

  50. [50]

    Anqi Xiao, Weichen Yu, and Hongyuan Yu. 2025. Sample-Aware RandAugment: Search-Free Automatic Data Augmentation for Effective Image Recognition: A. ExDet: Open-Domain Open-Vocabulary Detection with Cross-modal Extrapolation and Rectification MM ’26, November 10–14, 2026, Janeiro, Brazil Xiao et al.International Journal of Computer Vision133, 11 (2025), 7710–7725

    2025

  51. [51]

    Mingjun Xu, Lingyun Qin, Weijie Chen, Shiliang Pu, and Lei Zhang. 2023. Multi- view adversarial discriminator: Mine the non-causal factors for object detection in unseen domains. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 8103–8112

    2023

  52. [52]

    Shilin Xu, Xiangtai Li, Size Wu, Wenwei Zhang, Yunhai Tong, and Chen Change Loy. 2024. DST-Det: Open-Vocabulary Object Detection via Dynamic Self- Training.IEEE Transactions on Circuits and Systems for Video Technology(2024)

    2024

  53. [53]

    Xiaoran Xu, Jiangang Yang, Wenhui Shi, Siyuan Ding, Luqing Luo, and Jian Liu

  54. [54]

    InProceedings of the AAAI Conference on Artificial Intelligence

    PhysAug: A Physical-guided and Frequency-based Data Augmentation for Single-Domain Generalized Object Detection. InProceedings of the AAAI Conference on Artificial Intelligence. 21815–21823

  55. [55]

    Yifan Xu, Mengdan Zhang, Xiaoshan Yang, and Changsheng Xu. 2023. Exploring multi-modal contextual knowledge for open-vocabulary object detection.arXiv preprint arXiv:2308.15846(2023)

    work page arXiv 2023

  56. [56]

    Alireza Zareian, Kevin Dela Rosa, Derek Hao Hu, and Shih-Fu Chang. 2021. Open-vocabulary object detection using captions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 14393–14402

    2021

  57. [57]

    Haozhuo Zhang, Huimin Yu, Yuming Yan, and Runfa Wang. 2022. Gated domain- invariant feature disentanglement for domain generalizable object detection. arXiv preprint arXiv:2203.11432(2022)

    work page arXiv 2022

  58. [58]

    Yupeng Zhang, Ruize Han, Zhiwei Chen, Wei Feng, and Liang Wan. 2026. NoOVD: Novel Category Discovery and Embedding for Open-Vocabulary Object Detection. arXiv:2603.21069 [cs.CV] https://arxiv.org/abs/2603.21069

    work page arXiv 2026

  59. [59]

    Yupeng Zhang, Ruize Han, Fangnan Zhou, Song Wang, Wei Feng, and Liang Wan. 2025. ODOV: Towards Open-Domain Open-Vocabulary Object Detection. Preprint submitted to arXiv on August 2, 2025; expected to be available shortly

    2025

  60. [60]

    Yupeng Zhang, Shuqi Zheng, Ruize Han, Yuzhong Feng, Junhui Hou, Linqi Song, Wei Feng, and Liang Wan. [n. d.]. Rethinking the One-shot Object Detection: Cross-Domain Object Search. InACM Multimedia 2024

    2024

  61. [61]

    Shiyu Zhao, Samuel Schulter, Long Zhao, Zhixing Zhang, Yumin Suh, Man- mohan Chandraker, Dimitris N Metaxas, et al. 2024. Taming self-training for open-vocabulary object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 13938–13947

    2024

  62. [62]

    Yiwu Zhong, Jianwei Yang, Pengchuan Zhang, Chunyuan Li, Noel Codella, Liu- nian Harold Li, Luowei Zhou, Xiyang Dai, Lu Yuan, Yin Li, et al. 2022. Regionclip: Region-based language-image pretraining. InProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition. 16793–16803

    2022

  63. [63]

    Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xiang. 2024. Mixstyle neural networks for domain generalization and adaptation.International Journal of Computer Vision132, 3 (2024), 822–836

    2024

  64. [64]

    Xingyi Zhou, Rohit Girdhar, Armand Joulin, Philipp Krähenbühl, and Ishan Misra. 2022. Detecting twenty-thousand classes using image-level supervision. InEuropean Conference on Computer Vision. 350–368

    2022