Pith. sign in

REVIEW 4 major objections 5 minor 124 references

TITAN: Query-Token based Domain Adaptive Adversarial Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A variance-based split of the target domain plus query-token adversarial alignment is claimed to push source-free object detection to new state of the art, with mAP gains up to +22.7 over prior methods.

desk verdict TITAN's central variance-based target split is vacuous under its own dropout=0.0 setting, and the headline gains do not match the tables; the method as written cannot run as an SF-DAOD method. read the letter →

arxiv 2506.21484 v1 pith:YRX2PSET submitted 2025-06-26 cs.CV cs.AI

Tajamul Ashraf , Janibul Bashir This is my paper
classification cs.CVcs.AI
keywords source-freedomainadaptationobjectdetectionadversariallearningquery-tokenvariance-basedpartitioningstudent-teacherFocalNet-DINObreastcancer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that the biggest obstacle in source-free domain-adaptive object detection — pseudo-labels going bad when the teacher collapses — can be removed by splitting the unlabeled target images into two piles: those that look like the source and those that don't. It proposes to do the split by ranking images by the variance of Monte-Carlo dropout predictions, asserting that high variance marks source-similar images. The two piles are then aligned to each other and to the model with adversarial losses attached to the query and token embeddings of a transformer detector, inside a student-teacher loop. On four natural-image benchmarks the paper reports mAP gains of +22.7, +22.2, +21.1, and +3.7 over previous state of the art, and on two mammography benchmarks recall improvements from 0.25 to 0.78 and 0.15 to 0.51 at 0.3 false positives per image. If correct, this would make source-free adaptation practical for real deployments where source images cannot be shared.

What carries the argument

The central object is the detection variance $v_i = v_{b_i} v_{c_i}$, the product of a bounding-box coordinate variance and a class-score variance estimated from $M$ stochastic forward passes under dropout; ranking target images by this variance and thresholding at a rank fraction $\sigma$ divides the target into easy and hard subsets. Around this, the method adds adversarial learning to a transformer detector: a domain query is concatenated to the encoder and decoder token sequences and passed through domain discriminators with a domain label $d$, and each token is also passed through per-token discriminators, all trained through gradient reversal so the detector learns domain-invariant representations. A cascaded loss sums these query and token losses over encoder and decoder layers with balancing coefficients $\lambda_{encq}$ and $\lambda_{decq}$.

What would settle it

Run the released training code with the target-division ranks randomly permuted (easy/hard labels shuffled) and check whether the mAP gain persists; or inspect whether the domain discriminator's $d=0$ branch receives real source images during adaptation, which would break the claim of a strictly source-free setting.

Watch

Extended reading notes

Core claim

The paper introduces TITAN, a source-free domain adaptive object detection method built on the claim that the target domain can be partitioned into source-similar (easy) and source-dissimilar (hard) subsets by ranking images with a detection variance computed from Monte-Carlo dropout forward passes. It further claims that the teacher-student training instability, caused by noisy pseudo-labels under domain shift, can be suppressed by adversarially aligning features at both the query level and the per-token level inside a FocalNet-DINO transformer detector. The reported result is state-of-the-art performance on four natural-image benchmarks (C2F, C2B, S2C, K2C) and on two cross-dataset breast-cancer detection tasks, with the largest gains appearing where prior source-free methods were weakest.

Load-bearing premise

The adversarial losses need two labeled domains ($d=0$ for source and $d=1$ for target), so the method secretly assumes either that source images are available after all, or that the high-variance 'easy' half of the target data behaves like the source domain well enough to stand in for it; neither is established.

Editorial extensions

If this is right

  • If the reported gains hold, source-free detection can be solved by partitioning the target domain rather than by accessing source data or source-feature statistics.
  • The query-token adversarial module is a plug-in for transformer detectors, so it could transfer to other DETR-style architectures beyond FocalNet-DINO.
  • The medical results suggest that cross-institution and cross-machine breast cancer screening can work without sharing patient data, provided the variance-similarity link holds on clinical images.
  • The large margins over prior SF-DAOD methods on C2F, C2B, and S2C imply that target division plus adversarial alignment can rescue student-teacher pipelines from early collapse.

Reading between the lines

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

  • The variance ranking may be picking up pseudo-label noise or detection difficulty rather than true source-similarity; the paper does not separate these explanations, and a test using oracle source-feature distances would clarify which one drives the gains.
  • Because the adversarial losses explicitly use source-labeled samples ($d=0$), a strictly source-free implementation would need to generate or substitute source-like features from the easy subset; this is an engineering step the paper leaves implicit.
  • The medical recall jumps (0.25 to 0.78, 0.15 to 0.51) are far larger than the natural-image gains, hinting that the partition may be exploiting small-target count or class-imbalance effects rather than pure domain alignment; per-image false-positive analysis would test this.
  • A direct comparison against random assignment of easy/hard labels, or against a confidence-based (low-score) split, would isolate whether the variance criterion itself carries the benefit or whether any two-group split helps the adversarial objective.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. TITAN proposes a source-free domain-adaptive object detection (SF-DAOD) method built on FocalNet-DINO. The method first partitions the unlabeled target domain into source-similar (easy) and source-dissimilar (hard) subsets using a Monte-Carlo-dropout estimate of detection variance, and then applies query-token and token-wise adversarial losses inside a student-teacher framework to align the two groups. The paper reports large improvements over prior SF-DAOD and UDA methods on four natural-image benchmarks (C2F, C2B, S2C, K2C) and two medical benchmarks (RSNA-to-INBreast and DDSM-to-INBreast), with ablations attributing much of the gain to the target division and adversarial components.

Significance. If the method worked as described, TITAN would be a significant advance: the variance-based target partition is a novel and potentially useful mechanism for source-free detection, the reported gains over strong baselines such as PETS and A2SFOD are substantial, and the inclusion of medical imaging benchmarks addresses a relevant privacy-constrained application. The paper also contains reproducibility-friendly elements: explicit hyperparameter tables, augmentation details, and a supplemental theoretical derivation. However, the central mechanisms as written are not operational. The MC-dropout variance is identically zero under the paper's own dropout=0.0 configuration, the adversarial losses require source-domain labels that the source-free setting forbids, and the medical experiments train and test on the same target images. These issues are load-bearing rather than cosmetic, so the empirical and conceptual claims cannot be accepted in their present form.

major comments (4)
  1. [Section 3.1, Eqs. (2)-(3); Supplementary Table 7] The target partition mechanism is degenerate under the reported configuration. Section 3.1 defines the detection variance via M Monte-Carlo forward passes 'using dropout,' but Supplementary Table 7 lists dropout: 0.0 for the FocalNet-DINO configuration. With dropout probability zero, every forward pass is deterministic, so b^m_ij = mean(b_ij) and c^m_ij = mean(c_ij) for all m, giving v_b_i = v_c_i = v_i = 0 for every image. The rank r_i is then arbitrary and the threshold sigma no longer separates source-similar from source-dissimilar images. Since Table 5 and the abstract attribute a large part of the reported improvement to this target division, the central causal mechanism collapses.
  2. [Section 3.2, Eqs. (5),(7); Section 3.3, Eqs. (8)-(9); Fig. 2] The adversarial losses require two labeled domains, but the paper declares the setting source-free. Equations (5), (7), (8), and (9) use a domain label d taking value 0 for source images and 1 for target images, and Fig. 2 shows 'Source Data' entering the pipeline. The manuscript states in the abstract and Section 3 that source data is unavailable during adaptation. If the intended d=0 class is instead the easy target subset used as a source proxy, that substitution is nowhere stated, derived, or validated. As written, the method either assumes source data that the problem definition forbids or, in the proxy interpretation, aligns the target to itself rather than to the source.
  3. [Supplementary Section B.2, Table 6] The medical benchmarks are evaluated on the same images used for adaptation. The supplementary note explicitly says that for medical datasets the whole target dataset is used 'during training and testing,' and Table 6 confirms that for RSNA-BSD1K, INBreast, and DDSM the target train and test sets have identical sizes (e.g., 410/410 for INBreast and 1000/1000 for RSNA-BSD1K). This means the reported FROC curves and AUC values on R2In and D2In are not measured on unseen target data, so the SOTA claims on these benchmarks are not supported.
  4. [Section 3.1, Fig. 1] The stated intuition for the variance-based partition is internally contradictory. The text says 'A higher variance suggests a stronger resemblance to the source domain' but also 'the model exhibits greater uncertainty (hard samples) when predicting source-similar images, leading to elevated variance values.' These two statements assign opposite meanings to high variance: one calls high-variance images easy/source-similar, the other calls them hard. No evidence is provided that detection variance is monotonically related to source-domain resemblance, and the paper's own description does not specify which notion of 'hard' is meant.
minor comments (5)
  1. [Section 1 and Table 1] The reported C2F mAP is inconsistent: Section 1 lists 50.2 for TITAN, while Table 1 reports 52.2; the abstract's improvement figures (+22.7, +22.2, +21.1, +3.7) appear to be relative percentages computed from the stated absolute numbers, but they are not labeled as such.
  2. [Table 5] The ablation table is difficult to parse: the column headers and checkmarks do not make it clear which rows correspond to query-based versus token-wise losses, and the text refers to 'TITAN applied to the last encoder and decoder layers' without defining which acronym denotes the full method.
  3. [Fig. 4 and Fig. 7] The caption of Fig. 4 labels the method 'SFA (Ours)' although the paper's method is TITAN, and the caption of Fig. 7 says 'Quantitative results' when the content is qualitative; these labels should be corrected.
  4. [Supplementary Section A] The theoretical proof is titled 'Proof of Preposition' and uses a five-layer decomposition while Section 3.4 states that the discriminators are three-layer MLPs; the proof should be reconciled with the architecture actually used.
  5. [References] Several references are incomplete or garbled, e.g., reference [7] has a corrupted author list, and some entries lack venue or page information; a careful copyedit is needed.

Circularity Check

2 steps flagged · score 8.0 of 10

TITAN's variance-defined 'source-similar' subset makes its adversarial alignment a self-alignment of target data, and the MC-dropout variance is identically zero under the paper's own dropout=0 setting.

  1. self definitional [Section 3.1 (target division), Section 3.2 Eq. (5), Fig. 2]
    "Even though the source data is inaccessible, the source-pretrained model retains crucial knowledge about the source domain... If vl i ≥ σ, we categorize Xi as source-similar; otherwise, it is deemed source-dissimilar. ... d is the domain label, taking the value 0 for source images and 1 for target images."

    The source-similar and source-dissimilar labels are assigned by the variance threshold (vl_i ≥ σ), so 'source similarity' is not measured against the actual source; it is definitionally whatever has high variance under the source-pretrained model. The adversarial loss that supposedly closes the source-target gap is written with d=0 for 'source images' and d=1 for 'target images', but in the declared source-free setting the only candidates for d=0 are the paper's self-defined easy target images (or, per Fig. 2, source data that the source-free setting forbids). In either horn, the alignment is between two partitions of target data, so the claimed source-to-target transfer reduces to target self-alignment.

  2. other [Section 3.1, Eqs. (2)-(3); Supplementary Table 7]
    "This approximation is achieved by conducting M stochastic forward passes while keeping the detection model unchanged [5]. ... dropout | Dropout | 0.0"

    Equations (2)-(3) define v_bi and v_ci as the mean squared deviation of M Monte-Carlo forward passes. With dropout probability 0.0, every forward pass has identical weights and activations, so b^m_ij = \bar b_ij and c^m_ij = \bar c_ij for all m, making v_bi = v_ci = v_i = 0 for every image. The rank r_i is then arbitrary, and the threshold σ cannot separate source-similar from source-dissimilar images. The paper's core variance-to-source-similarity mechanism therefore produces no easy/hard split under its own hyperparameters; any gain credited to target division cannot be traced to the proposed mechanism. This is a by-construction degeneracy of the central derivation chain.

full rationale

The paper's main claimed derivation is: variance of MC-dropout predictions partitions the target into source-similar (easy) and source-dissimilar (hard) subsets; query-token adversarial losses then align these subsets to transfer from source to target. That chain is circular or vacuous at its two load-bearing joints. First, 'source-similar' is defined by the very variance criterion that is supposed to diagnose source similarity, and the adversarial losses, which formally require d=0 source images, can only be executed in the source-free setting by using the easy target subset as the source side; the alignment is therefore target-to-target, not source-to-target. Second, the variance computation itself is degenerate under the paper's own configuration: Supplementary Table 7 sets dropout to 0.0, so the Monte-Carlo passes are deterministic, forcing all variances to zero and making the easy/hard partition arbitrary. The reported mAP numbers are externally benchmarked against other methods, so those numbers are not circular in themselves; what is circular is the paper's attribution of those numbers to TITAN's proposed mechanism. The generalization bound is imported from prior work [2,33] but is not a self-citation and does not bear on the core adaptation claim. No load-bearing self-citation chain is present; the circularity is internal to the method's own definitions and hyperparameters.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The main cost of the method is the variance-to-source-similarity premise, which is asserted rather than derived, plus the implicit assumption that a source-like domain is available for adversarial training despite the source-free framing. The implementation also silently assumes a nonzero dropout rate for MC sampling and a specified partition threshold, neither of which is provided.

free parameters (3)
  • sigma (target division threshold) = not reported
    Rank-fraction threshold in Section 3.1 that decides source-similar vs source-dissimilar. It is load-bearing for the entire easy/hard split, yet no value, no sensitivity analysis, and no tuning procedure are given.
  • M (number of Monte Carlo dropout passes) = not reported
    Eqs. 2-3 require M stochastic forward passes to estimate variance. M is never specified, and the reported dropout rate of 0.0 makes the stochasticity assumption vacuous.
  • lambda_encq and lambda_decq = 0.1 and 0.1
    Balancing weights in Eqs. 10-11 for query-based vs token-based adversarial losses. Hand-chosen hyperparameters with no reported sensitivity study.
assumptions (4)
  • ad hoc to paper Higher detection variance implies higher recall and stronger source-domain resemblance.
    Core premise of Section 3.1 and Fig. 1. Presented as an insight with no derivation, no empirical validation, and the text itself labels source-similar images as 'hard samples', blurring the direction of the effect.
  • domain assumption A domain discriminator can be trained with source and target labels in a source-free setting.
    Eqs. 5, 7-9 use d=0 for source and d=1 for target; Figure 2 shows 'Source Data' flowing into the pipeline, conflicting with the SF-DAOD premise that source data is unavailable.
  • domain assumption Monte Carlo dropout approximates predictive variance for the detector.
    Section 3.1 invokes Gal and Ghahramani, but Table 7 sets dropout=0.0, so the approximation is vacuous as configured.
  • standard math Standard spectral-norm covering bounds for MLP discriminators.
    Section 3.6 Theorem cites Bartlett et al. The supplement proof uses a 5-layer product while the theorem states 3 layers, and the bound is not connected to TITAN's actual pipeline.
invented entities (1)
  • Domain query token (q_enc^d, q_dec^d)
    purpose: Appended to encoder/decoder sequences to aggregate domain-specific context for adversarial alignment (Eqs. 4 and 6).
    New learnable component. Evaluated only within the paper's own ablations, with no external falsifiable handle or cross-check outside the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TITAN: Query-Token based Domain Adaptive Adversarial Learning." pith.science (2026). https://pith.science/paper/YRX2PSET

@misc{pith2026250621484,
  author       = {Pith},
  title        = {Pith review of: TITAN: Query-Token based Domain Adaptive Adversarial Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YRX2PSET}},
  note         = {Machine review of arXiv:2506.21484}
}
read the original abstract

We focus on the source-free domain adaptive object detection (SF-DAOD) problem when source data is unavailable during adaptation and the model must adapt to an unlabeled target domain. The majority of approaches for the problem employ a self-supervised approach using a student-teacher (ST) framework where pseudo-labels are generated via a source-pretrained model for further fine-tuning. We observe that the performance of a student model often degrades drastically, due to the collapse of the teacher model, primarily caused by high noise in pseudo-labels, resulting from domain bias, discrepancies, and a significant domain shift across domains. To obtain reliable pseudo-labels, we propose a Target-based Iterative Query-Token Adversarial Network (TITAN), which separates the target images into two subsets: those similar to the source (easy) and those dissimilar (hard). We propose a strategy to estimate variance to partition the target domain. This approach leverages the insight that higher detection variances correspond to higher recall and greater similarity to the source domain. Also, we incorporate query-token-based adversarial modules into a student-teacher baseline framework to reduce the domain gaps between two feature representations. Experiments conducted on four natural imaging datasets and two challenging medical datasets have substantiated the superior performance of TITAN compared to existing state-of-the-art (SOTA) methodologies. We report an mAP improvement of +22.7, +22.2, +21.1, and +3.7 percent over the current SOTA on C2F, C2B, S2C, and K2C benchmarks, respectively.

Figures

Figures reproduced from arXiv: 2506.21484 by the authors.

Figure 1
Figure 1. The core idea of our framework (TITAN) is that higher detection variances signal high recall and similarity to the source domain, enabling us to divide the target domain into easy and hard subsets. vised training of deep object detection models. Unsupervised Domain Adaptation (UDA). It has been widely observed that, despite their effectiveness in familiar visual contexts, deep object detection models often struggle … view at source ↗
Figure 2
Figure 2. Overview: Target-based Iterative Query-Token Adversarial Network ( [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. "Visualization of domain queries from both the encoder and decoder sides of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative results comparing Source model [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results for car detection on Cityscapes in [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Qualitative results for car detection on Foggy Cityscapes in [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Quantitative results for Breast Cancer detection in [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

124 extracted references · 72 canonical work pages

  1. [1]

    Attention is all you need

    Vaswani Ashish. Attention is all you need. Advances in neural information processing systems, 30:I, 2017. 2

    2017

  2. [2]

    Spectrally-normalized margin bounds for neural networks

    Peter L Bartlett, Dylan J Foster, and Matus J Telgarsky. Spectrally-normalized margin bounds for neural networks. Advances in neural information processing systems , 30,

  3. [3]

    Weakly supervised deep detection networks

    Hakan Bilen and Andrea Vedaldi. Weakly supervised deep detection networks. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 2846– 2854, 2016. 2

    2016

  4. [4]

    Weakly supervised object detection with convex clustering

    Hakan Bilen, Marco Pedersoli, and Tinne Tuytelaars. Weakly supervised object detection with convex clustering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1081–1089, 2015. 2

    2015

  5. [5]

    Weight uncertainty in neural networks

    Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural networks. In ICML, page 1613–1622, 2015. 3

    2015

  6. [6]

    End- to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End- to-end object detection with transformers. In European conference on computer vision , pages 213–229. Springer,

  7. [7]

    Rsna screening mammog- raphy breast cancer detection

    C Carr, F Kitamura, J K-Cramer, J Mongan, K Andriole, M V , M Riopel, R Ball, and S Dane. Rsna screening mammog- raphy breast cancer detection. 2022. 2, 6, 8, 3

    2022

  8. [8]

    Harmonizing transferability and discriminability for adapting object detectors

    Chaoqi Chen, Zebiao Zheng, Xinghao Ding, Yue Huang, and Qi Dou. Harmonizing transferability and discriminability for adapting object detectors. In CVPR, pages 8866–8875,

Show all 124 references
  1. [9]

    Domain adaptive faster r-cnn for object detection in the wild

    Yuhua Chen, Wen Li, Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Domain adaptive faster r-cnn for object detection in the wild. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. 6, 7

    2018

  2. [10]

    Domain adaptive faster r-cnn for object detection in the wild

    Yuhua Chen, Wen Li, Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Domain adaptive faster r-cnn for object detection in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 3339– 3348, 2018. 1

    2018

  3. [11]

    No more discrim- ination: Cross city adaptation of road scene segmenters

    Yi-Hsin Chen, Wei-Yu Chen, Yu-Ting Chen, Bo-Cheng Tsai, Yu-Chiang Frank Wang, and Min Sun. No more discrim- ination: Cross city adaptation of road scene segmenters. In Proceedings of the IEEE International Conference on Computer Vision, pages 1992–2001, 2017. 1

    1992

  4. [12]

    Towards large-scale small object detection: Survey and benchmarks

    Gong Cheng, Xiang Yuan, Xiwen Yao, Kebing Yan, Qinghua Zeng, Xingxing Xie, and Junwei Han. Towards large-scale small object detection: Survey and benchmarks. IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2023. 2

    2023

  5. [13]

    Adversarial alignment for source free object detection

    Qiaosong Chu, Shuyan Li, Guangyi Chen, Kai Li, and Xiu Li. Adversarial alignment for source free object detection. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 452–460, 2023. 1, 2, 3, 6, 7

    2023

  6. [14]

    Weakly supervised object localization with multi- fold multiple instance learning

    Ramazan Gokberk Cinbis, Jakob Verbeek, and Cordelia Schmid. Weakly supervised object localization with multi- fold multiple instance learning. IEEE transactions on pattern analysis and machine intelligence, 39(1):189–203,

  7. [15]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In Proceed- ings of the IEEE conference on computer vision and pattern re...

    2016

  8. [16]

    Up-detr: Unsupervised pre-training for object detection with transformers

    Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object detection with transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 1601– 1610, 2021. 2

    2021

  9. [17]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 2

    2009

  10. [18]

    Un- biased mean teacher for cross-domain object detection

    Jinhong Deng, Wen Li, Yuhua Chen, and Lixin Duan. Un- biased mean teacher for cross-domain object detection. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR) , pages 4091–4101,

  11. [19]

    Un- biased mean teacher for cross-domain object detection

    Jinhong Deng, Wen Li, Yuhua Chen, and Lixin Duan. Un- biased mean teacher for cross-domain object detection. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 4091–4101, 2021. 2, 7

    2021

  12. [20]

    Harmo- nious teacher for cross-domain object detection

    Jinhong Deng, Dongli Xu, Wen Li, and Lixin Duan. Harmo- nious teacher for cross-domain object detection. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23829–23838, 2023. 6

    2023

  13. [21]

    Confident anchor-induced multi-source free domain adaptation

    Jiahua Dong, Zhen Fang, Anjin Liu, Gan Sun, and Tongliang Liu. Confident anchor-induced multi-source free domain adaptation. Advances in Neural Information Processing Systems, 34:2848–2860, 2021. 1

    2021

  14. [22]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint a...

    2010 arXiv

  15. [23]

    The pascal visual object classes (voc) challenge

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. International journal of computer vision, 88:303–338, 2010. 1, 2

    2010

  16. [24]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In ICML, pages 1050–1059, 2016. 3

    2016

  17. [25]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pas- cal Germain, Hugo Larochelle, François Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks. Journal of machine learning research, 17(59):1–35, 2016. 1, 6

    2016

  18. [26]

    Csda: Learning category-scale joint feature for domain adaptive object detection

    Changlong Gao, Chengxu Liu, Yujie Dun, and Xueming Qian. Csda: Learning category-scale joint feature for domain adaptive object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11421– 11430, 2023. 6, 7

    2023

  19. [27]

    Vision meets robotics: The kitti dataset

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset. The International Journal of Robotics Research, 32(11):1231– 1237, 2013. 1, 5, 2

    2013

  20. [28]

    Fast r-cnn

    Ross Girshick. Fast r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 1440– 1448, 2015. 2

    2015

  21. [29]

    Improving transferability for domain adaptive detection transformers

    Kaixiong Gong, Shuang Li, Shugang Li, Rui Zhang, Chi Harold Liu, and Qiang Chen. Improving transferability for domain adaptive detection transformers. In Proceedings of the 30th ACM International Conference on Multimedia, pages 1543–1551, 2022. 6, 7

    2022

  22. [30]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014. 5

    2014

  23. [31]

    Boosting weakly supervised object detection using fusion and priors from hallucinated depth

    Cagri Gungor and Adriana Kovashka. Boosting weakly supervised object detection using fusion and priors from hallucinated depth. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 739– 748, 2024. 2

    2024

  24. [32]

    Recent advances in deep learning theory

    Fengxiang He and Dacheng Tao. Recent advances in deep learning theory. arXiv preprint arXiv:2012.10931, 2020. 5

    2012 arXiv

  25. [33]

    Why resnet works? residuals generalize

    Fengxiang He, Tongliang Liu, and Dacheng Tao. Why resnet works? residuals generalize. IEEE transactions on neural networks and learning systems, 31(12):5349–5362, 2020. 5

    2020

  26. [34]

    Cross domain object detection by target-perceived dual branch distillation

    Mengzhe He, Yali Wang, Jiaxi Wu, Yiru Wang, Hanqing Li, Bo Li, Weihao Gan, Wei Wu, and Yu Qiao. Cross domain object detection by target-perceived dual branch distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9570–9580,

  27. [35]

    Cross domain object detection by target-perceived dual branch distillation

    Mengzhe He, Yali Wang, Jiaxi Wu, Yiru Wang, Hanqing Li, Bo Li, Weihao Gan, Wei Wu, and Yu Qiao. Cross domain object detection by target-perceived dual branch distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9570–9580, 2022. 6, 7

    2022

  28. [36]

    Multi-adversarial faster-rcnn for unrestricted object detection

    Zhenwei He and Lei Zhang. Multi-adversarial faster-rcnn for unrestricted object detection. In Proceedings of the IEEE/CVF international conference on computer vision , pages 6668–6677, 2019. 1

    2019

  29. [37]

    Fcns in the wild: Pixel-level adversarial and constraint-based adaptation

    Judy Hoffman, Dequan Wang, Fisher Yu, and Trevor Darrell. Fcns in the wild: Pixel-level adversarial and constraint-based adaptation. arXiv preprint arXiv:1612.02649, 2016. 1

    2016 arXiv

  30. [38]

    Cycada: Cycle-consistent adversarial domain adaptation

    Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei Efros, and Trevor Darrell. Cycada: Cycle-consistent adversarial domain adaptation. In ICML, pages 1989–1998, 2018. 1

    1989

  31. [39]

    Every pixel matters: Center-aware feature alignment for domain adaptive object detector

    Cheng-Chun Hsu, Yi-Hsuan Tsai, Yen-Yu Lin, and Ming- Hsuan Yang. Every pixel matters: Center-aware feature alignment for domain adaptive object detector. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IX 16 , pages 73...

    2020

  32. [40]

    Model adaptation: Historical contrastive learning for unsu- pervised domain adaptation without source data

    Jiaxing Huang, Dayan Guan, Aoran Xiao, and Shijian Lu. Model adaptation: Historical contrastive learning for unsu- pervised domain adaptation without source data. Advances in Neural Information Processing Systems, 34:3635–3649,

  33. [41]

    Model adaptation: Historical contrastive learning for un- supervised domain adaptation without source data, 2022

    Jiaxing Huang, Dayan Guan, Aoran Xiao, and Shijian Lu. Model adaptation: Historical contrastive learning for un- supervised domain adaptation without source data, 2022. 6

    2022

  34. [42]

    Aqt: Adversarial query transformers for domain adaptive object detection

    Wei-Jie Huang, Yu-Lin Lu, Shih-Yao Lin, Yusheng Xie, and Yen-Yu Lin. Aqt: Adversarial query transformers for domain adaptive object detection. In IJCAI, pages 972–979, 2022. 1, 6, 7

    2022

  35. [43]

    Cross-domain weakly-supervised object detection through progressive domain adaptation

    Naoto Inoue, Ryosuke Furuta, Toshihiko Yamasaki, and Kiyoharu Aizawa. Cross-domain weakly-supervised object detection through progressive domain adaptation. In Pro- ceedings of the IEEE conference on computer vision and pattern recognition, pages 5001–5009, 2018. 1

    2018

  36. [44]

    Decoupled adaptation for cross-domain object detection, 2022

    Junguang Jiang, Baixu Chen, Jianmin Wang, and Ming- sheng Long. Decoupled adaptation for cross-domain object detection, 2022. 6, 7

    2022

  37. [45]

    Variational model perturbation for source-free do- main adaptation

    Mengmeng Jing, Xiantong Zhen, Jingjing Li, and Cees Snoek. Variational model perturbation for source-free do- main adaptation. Advances in Neural Information Process- ing Systems, 35:17173–17187, 2022. 1

    2022

  38. [46]

    Driving in the matrix: Can virtual worlds replace human- generated annotations for real world tasks? arXiv preprint arXiv:1610.01983, 2016

    Matthew Johnson-Roberson, Charles Barto, Rounak Mehta, Sharath Nittur Sridhar, Karl Rosaen, and Ram Vasudevan. Driving in the matrix: Can virtual worlds replace human- generated annotations for real world tasks? arXiv preprint arXiv:1610.01983, 2016. 5, 8, 2

    arXiv

  39. [47]

    Haupt- mann

    Guoliang Kang, Lu Jiang, Yi Yang, and Alexander G. Haupt- mann. Contrastive adaptation network for unsupervised domain adaptation. In CVPR, pages 4893–4902, 2019. 3

    2019

  40. [48]

    To miss-attend is to misalign! residual self-attentive feature alignment for adapting object detectors

    Vaishnavi Khindkar, Chetan Arora, Vineeth N Balasubrama- nian, Anbumani Subramanian, Rohit Saluja, and CV Jawa- har. To miss-attend is to misalign! residual self-attentive feature alignment for adapting object detectors. In Proceed- ings of the IEEE/CVF Winter Conference on Ap...

    2022

  41. [49]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980,

    arXiv

  42. [50]

    Weakly supervised virus capsid detection with image-level annotations in electron microscopy images

    Hannah Kniesel, Leon Sick, Tristan Payer, Tim Bergner, Kavitha Shaga Devan, Clarissa Read, Paul Walther, Timo Ropinski, and Pedro Hermosilla. Weakly supervised virus capsid detection with image-level annotations in electron microscopy images. In The Twelfth International Confe...

    2023

  43. [51]

    A curated mammography data set for use in computer-aided detection and diagnosis research

    Rebecca Sawyer Lee, Francisco Gimenez, Assaf Hoogi, Kanae Kawai Miyake, Mia Gorovoy, and Daniel L Rubin. A curated mammography data set for use in computer-aided detection and diagnosis research. Scientific data, 4(1):1–9,

  44. [52]

    Spatial attention pyramid network for unsupervised domain adaptation, 2020

    Congcong Li, Dawei Du, Libo Zhang, Longyin Wen, Tiejian Luo, Yanjun Wu, and Pengfei Zhu. Spatial attention pyramid network for unsupervised domain adaptation, 2020. 7

    2020

  45. [53]

    Source-free object detection by learning to overlook domain style

    Shuaifeng Li, Mao Ye, Xiatian Zhu, Lihua Zhou, and Lin Xiong. Source-free object detection by learning to overlook domain style. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8014– 8023, 2022. 2, 3, 6, 7

    2022

  46. [54]

    Sigma: Semantic- complete graph matching for domain adaptive object de- tection

    Wuyang Li, Xinyu Liu, and Yixuan Yuan. Sigma: Semantic- complete graph matching for domain adaptive object de- tection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5291–5300, 2022. 6

    2022

  47. [55]

    A free lunch for unsuper- vised domain adaptive object detection without source data,

    Xianfeng Li, Weijie Chen, Di Xie, Shicai Yang, Peng Yuan, Shiliang Pu, and Yueting Zhuang. A free lunch for unsuper- vised domain adaptive object detection without source data,

  48. [56]

    A free lunch for unsu- pervised domain adaptive object detection without source data

    Xianfeng Li, Weijie Chen, Di Xie, Shicai Yang, Peng Yuan, Shiliang Pu, and Yueting Zhuang. A free lunch for unsu- pervised domain adaptive object detection without source data. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 8474–8481, 2021. 2, 3, 6

    2021

  49. [57]

    Cross-domain adaptive teacher for object detection

    Yu-Jhe Li, Xiaoliang Dai, Chih-Yao Ma, Yen-Cheng Liu, Kan Chen, Bichen Wu, Zijian He, Kris Kitani, and Peter Vajda. Cross-domain adaptive teacher for object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7581–7590, 2022. 6

    2022

  50. [58]

    Cross-domain adaptive teacher for object detection,

    Yu-Jhe Li, Xiaoliang Dai, Chih-Yao Ma, Yen-Cheng Liu, Kan Chen, Bichen Wu, Zijian He, Kris Kitani, and Peter Vajda. Cross-domain adaptive teacher for object detection,

  51. [59]

    Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation

    Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation. In International conference on machine learning, pages 6028–6039. PMLR, 2020. 1

    2020

  52. [60]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings...

    2014

  53. [61]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In ICCV, pages 2980–2988, 2017. 2

    2017

  54. [62]

    Periodically exchange teacher-student for source-free object detection

    Qipeng Liu, Luojun Lin, Zhifeng Shen, and Zhifeng Yang. Periodically exchange teacher-student for source-free object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6414–6424, 2023. 1, 2, 3, 6, 7

    2023

  55. [63]

    Source-free domain adaptation for semantic segmentation

    Yuang Liu, Wei Zhang, and Jun Wang. Source-free domain adaptation for semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1215–1224, 2021. 1

    2021

  56. [64]

    Cigar: Cross-modality graph reasoning for domain adaptive object detection

    Yabo Liu, Jinghua Wang, Chao Huang, Yaowei Wang, and Yong Xu. Cigar: Cross-modality graph reasoning for domain adaptive object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23776–23786, 2023. 6, 7

    2023

  57. [65]

    Learning feature decomposition for domain adaptive monocular depth esti- mation

    Shao-Yuan Lo, Wei Wang, Jim Thomas, Jingjing Zheng, Vishal M Patel, and Cheng-Hao Kuo. Learning feature decomposition for domain adaptive monocular depth esti- mation. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 8376–8382. IEEE, 2022. 1

    2022

  58. [66]

    Crots: Cross-domain teacher–student learning for source-free domain adaptive semantic segmen- tation

    Xin Luo, Wei Chen, Zhengfa Liang, Longqi Yang, Siwei Wang, and Chen Li. Crots: Cross-domain teacher–student learning for source-free domain adaptive semantic segmen- tation. International Journal of Computer Vision, 132(1): 20–39, 2024. 3

    2024

  59. [67]

    Codet: Co-occurrence guided region-word alignment for open-vocabulary object detection

    Chuofan Ma, Yi Jiang, Xin Wen, Zehuan Yuan, and Xiaojuan Qi. Codet: Co-occurrence guided region-word alignment for open-vocabulary object detection. Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  60. [68]

    Awada: Attention-weighted adversarial domain adaptation for object detection, 2022

    Maximilian Menke, Thomas Wenzel, and Andreas Schwung. Awada: Attention-weighted adversarial domain adaptation for object detection, 2022. 7

    2022

  61. [69]

    In- breast: toward a full-field digital mammographic database

    Inês C Moreira, Igor Amaral, Inês Domingues, António Cardoso, Maria Joao Cardoso, and Jaime S Cardoso. In- breast: toward a full-field digital mammographic database. Academic radiology, 19(2):236–248, 2012. 2, 6, 8

    2012

  62. [70]

    Ssal: Synergizing between self- training and adversarial learning for domain adaptive ob- ject detection

    Muhammad Akhtar Munir, Muhammad Haris Khan, M Sar- fraz, and Mohsen Ali. Ssal: Synergizing between self- training and adversarial learning for domain adaptive ob- ject detection. Advances in Neural Information Processing Systems, 34:22770–22782, 2021. 6

    2021

  63. [71]

    Unsupervised domain adap- tation of object detectors: A survey

    Poojan Oza, Vishwanath A Sindagi, Vibashan Vishnukumar Sharmini, and Vishal M Patel. Unsupervised domain adap- tation of object detectors: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023. 1, 2, 6

    2023

  64. [72]

    Automatic differentiation in pytorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Al- ban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. In NIPS-W, 2017. 6

    2017

  65. [73]

    Deep learning for detection of iso-dense, ob- scure masses in mammographically dense breasts

    Krithika Rangarajan, Pranjal Aggarwal, Dhruv Kumar Gupta, Rohan Dhanakshirur, Akhil Baby, Chandan Pal, Arun Kumar Gupta, Smriti Hari, Subhashis Banerjee, and Chetan Arora. Deep learning for detection of iso-dense, ob- scure masses in mammographically dense breasts. European ra...

    2023

  66. [74]

    Yolo9000: better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: better, faster, stronger. In Proceedings of the IEEE conference on com- puter vision and pattern recognition , pages 7263–7271,

  67. [75]

    You only look once: Unified, real-time object de- tection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016. 2

    2016

  68. [76]

    Faster r-cnn: Towards real-time object detection with re- gion proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with re- gion proposal networks. Advances in neural information processing systems, 28, 2015. 2

    2015

  69. [77]

    Seeking similarities over differences: Similarity-based do- main alignment for adaptive object detection

    Farzaneh Rezaeianaran, Rakshith Shetty, Rahaf Aljundi, Daniel Olmeda Reino, Shanshan Zhang, and Bernt Schiele. Seeking similarities over differences: Similarity-based do- main alignment for adaptive object detection. InProceedings of the IEEE/CVF International Conference on Co...

    2021

  70. [78]

    Foundations of machine learning, 2012

    Afshin Rostamizadeh, Ameet Talwalkar, and Mehryar Mohri. Foundations of machine learning, 2012. 5

    2012

  71. [79]

    Maximum classifier discrepancy for unsuper- vised domain adaptation

    Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku, and Tat- suya Harada. Maximum classifier discrepancy for unsuper- vised domain adaptation. In Proceedings of the IEEE con- ference on computer vision and pattern recognition, pages 3723–3732, 2018. 1

    2018

  72. [80]

    Strong-weak distribution alignment for adaptive object detection

    Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada, and Kate Saenko. Strong-weak distribution alignment for adaptive object detection. In CVPR, pages 6956–6965, 2019. 3

    2019

  73. [81]

    Strong-weak distribution alignment for adaptive object detection

    Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada, and Kate Saenko. Strong-weak distribution alignment for adaptive object detection. In Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition , pages 6956–6965, 2019. 1

    2019

  74. [82]

    Seman- tic foggy scene understanding with synthetic data

    Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Seman- tic foggy scene understanding with synthetic data. Interna- tional Journal of Computer Vision, 126:973–992, 2018. 5, 1, 2

    2018

  75. [83]

    Fully convolutional one-stage 3d ob- ject detection on lidar range images

    Zhi Tian, Xiangxiang Chu, Xiaoming Wang, Xiaolin Wei, and Chunhua Shen. Fully convolutional one-stage 3d ob- ject detection on lidar range images. Advances in Neural Information Processing Systems, 35:34899–34911, 2022. 2

    2022

  76. [84]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 7167–7176, 2017. 1

    2017

  77. [85]

    Sindagi, and Vishal M

    Vibashan VS, Vikram Gupta, Poojan Oza, Vishwanath A. Sindagi, and Vishal M. Patel. Mega-cda: Memory guided attention for category-aware unsupervised domain adaptive object detection, 2021. 7

    2021

  78. [86]

    Mixture of teacher experts for source-free do- main adaptive object detection

    Vibashan Vs, Poojan Oza, Vishwanath A Sindagi, and Vishal M Patel. Mixture of teacher experts for source-free do- main adaptive object detection. In 2022 IEEE International Conference on Image Processing (ICIP), pages 3606–3610. IEEE, 2022. 2

    2022

  79. [87]

    Instance relation graph guided source-free domain adaptive object detection

    Vibashan VS, Poojan Oza, and Vishal M Patel. Instance relation graph guided source-free domain adaptive object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3520–3530,

  80. [88]

    Towards on- line domain adaptive object detection

    Vibashan VS, Poojan Oza, and Vishal M Patel. Towards on- line domain adaptive object detection. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 478–488, 2023. 1

    2023

  81. [89]

    Weakly supervised lesion detection and diagnosis for breast cancers with partially annotated ultrasound images

    Jian Wang, Liang Qiao, Shichong Zhou, Jin Zhou, Jun Wang, Juncheng Li, Shihui Ying, Cai Chang, and Jun Shi. Weakly supervised lesion detection and diagnosis for breast cancers with partially annotated ultrasound images. IEEE Transac- tions on Medical Imaging, 2024. 2

    2024

  82. [90]

    Exploring sequence feature alignment for domain adaptive detection transformers

    Wen Wang, Yang Cao, Jing Zhang, Fengxiang He, Zheng- Jun Zha, Yonggang Wen, and Dacheng Tao. Exploring sequence feature alignment for domain adaptive detection transformers. In Proceedings of the 29th ACM International Conference on Multimedia, pages 1730–1738, 2021. 6, 7

    2021

  83. [91]

    Point linking network for object detection

    Xinggang Wang, Kaibing Chen, Zilong Huang, Cong Yao, and Wenyu Liu. Point linking network for object detection. arXiv preprint arXiv:1706.03646, 2017. 2

    2017 arXiv

  84. [92]

    Towards universal object detection by domain attention

    Xudong Wang, Zhaowei Cai, Dashan Gao, and Nuno Vas- concelos. Towards universal object detection by domain attention. In CVPR, pages 7289–7298, 2019. 3

    2019

  85. [93]

    Domain-specific suppres- sion for adaptive object detection, 2021

    Yu Wang, Rui Zhang, Shuo Zhang, Miao Li, YangYang Xia, XiShan Zhang, and ShaoLi Liu. Domain-specific suppres- sion for adaptive object detection, 2021. 7

    2021

  86. [94]

    Alwod: Active learning for weakly-supervised object detection

    Yuting Wang, Velibor Ilic, Jiatong Li, Branislav Kisaˇcanin, and Vladimir Pavlovic. Alwod: Active learning for weakly-supervised object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6459–6469, 2023. 2

    2023

  87. [95]

    Metateacher: Coordinating multi- model domain adaptation for medical image classification

    Zhenbin Wang, Mao Ye, Xiatian Zhu, Liuhan Peng, Liang Tian, and Yingying Zhu. Metateacher: Coordinating multi- model domain adaptation for medical image classification. Advances in Neural Information Processing Systems , 35: 20823–20837, 2022. 1

    2022

  88. [96]

    Adap- tive adversarial network for source-free domain adaptation

    Haifeng Xia, Handong Zhao, and Zhengming Ding. Adap- tive adversarial network for source-free domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9010–9019, 2021. 1

    2021

  89. [97]

    Source data-free domain adaptation of object detector through domain-specific perturbation

    Lin Xiong, Mao Ye, Dan Zhang, Yan Gan, Xue Li, and Yingying Zhu. Source data-free domain adaptation of object detector through domain-specific perturbation. International Journal of Intelligent Systems, 36, 2021. 6

    2021

  90. [98]

    Source data-free domain adaptation of object detector through domain-specific perturbation

    Lin Xiong, Mao Ye, Dan Zhang, Yan Gan, Xue Li, and Yingying Zhu. Source data-free domain adaptation of object detector through domain-specific perturbation. International Journal of Intelligent Systems, 36(8):3746–3766, 2021. 2, 3

    2021

  91. [99]

    Exploring categorical regularization for domain adap- tive object detection, 2020

    Chang-Dong Xu, Xing-Ran Zhao, Xin Jin, and Xiu-Shen Wei. Exploring categorical regularization for domain adap- tive object detection, 2020. 7

    2020

  92. [100]

    H2fa r-cnn: Holistic and hierarchical feature align- ment for cross-domain weakly supervised object detection

    Yunqiu Xu, Yifan Sun, Zongxin Yang, Jiaxu Miao, and Yi Yang. H2fa r-cnn: Holistic and hierarchical feature align- ment for cross-domain weakly supervised object detection. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14309–14319, 2022. 7

    2022

  93. [101]

    Focal modulation networks

    Jianwei Yang, Chunyuan Li, Xiyang Dai, and Jianfeng Gao. Focal modulation networks. Advances in Neural Information Processing Systems, 35:4203–4217, 2022. 1, 2, 3, 6

    2022

  94. [102]

    Exploiting the intrinsic neighborhood structure for source-free domain adaptation

    Shiqi Yang, Joost van de Weijer, Luis Herranz, Shangling Jui, et al. Exploiting the intrinsic neighborhood structure for source-free domain adaptation. Advances in neural informa- tion processing systems, 34:29393–29405, 2021. 1

    2021

  95. [103]

    At- tracting and dispersing: A simple approach for source-free domain adaptation

    Shiqi Yang, Shangling Jui, Joost van de Weijer, et al. At- tracting and dispersing: A simple approach for source-free domain adaptation. Advances in Neural Information Pro- cessing Systems, 35:5802–5815, 2022. 1

    2022

  96. [104]

    Enhancing source- free domain adaptive object detection with low-confidence pseudo label distillation

    Ilhoon Yoon, Hyeongjun Kwon, Jin Kim, Junyoung Park, Hyunsung Jang, and Kwanghoon Sohn. Enhancing source- free domain adaptive object detection with low-confidence pseudo label distillation. In European Conference on Com- puter Vision, pages 337–353. Springer, 2024. 6, 7

    2024

  97. [105]

    Bdd100k: A diverse driving dataset for heterogeneous multitask learning

    Fisher Yu, Haofeng Chen, Xin Wang, Wenqi Xian, Yingying Chen, Fangchen Liu, Vashisht Madhavan, and Trevor Dar- rell. Bdd100k: A diverse driving dataset for heterogeneous multitask learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition,...

    2020

  98. [106]

    Mttrans: Cross- domain object detection with mean teacher transformer

    Jinze Yu, Jiaming Liu, Xiaobao Wei, Haoyi Zhou, Yohei Nakata, Denis Gudovskiy, Tomoyuki Okuno, Jianxin Li, Kurt Keutzer, and Shanghang Zhang. Mttrans: Cross- domain object detection with mean teacher transformer. In European Conference on Computer Vision, pages 629–645. Spring...

    2022

  99. [107]

    Weakly supervised object detection using proposal-and semantic-level relationships

    Dingwen Zhang, Wenyuan Zeng, Jieru Yao, and Junwei Han. Weakly supervised object detection using proposal-and semantic-level relationships. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(6):3349–3363, 2020. 2

    2020

  100. [108]

    Weakly supervised object localization and detection: A survey

    Dingwen Zhang, Junwei Han, Gong Cheng, and Ming- Hsuan Yang. Weakly supervised object localization and detection: A survey. IEEE transactions on pattern analysis and machine intelligence, 44(9):5866–5885, 2021. 2

    2021

  101. [109]

    Dino: Detr with improved denoising anchor boxes for end-to-end object detection

    Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. In The Eleventh International Conference on Learning Representations, 2022. 1, 2

    2022

  102. [110]

    On the discrimination-generalization tradeoff in gans

    Pengchuan Zhang, Qiang Liu, Dengyong Zhou, Tao Xu, and Xiaodong He. On the discrimination-generalization tradeoff in gans. arXiv preprint arXiv:1711.02771, 2017. 5

    2017 arXiv

  103. [111]

    Divide and contrast: Source- free domain adaptation via adaptive contrastive learning

    Ziyi Zhang, Weikai Chen, Hui Cheng, Zhen Li, Siyuan Li, Liang Lin, and Guanbin Li. Divide and contrast: Source- free domain adaptation via adaptive contrastive learning. Advances in Neural Information Processing Systems , 35: 5137–5149, 2022. 1

    2022

  104. [112]

    Task-specific inconsistency alignment for domain adaptive object detection

    Liang Zhao and Limin Wang. Task-specific inconsistency alignment for domain adaptive object detection. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14217–14226, 2022. 6

    2022

  105. [113]

    Masked retraining teacher- student framework for domain adaptive object detection

    Zijing Zhao, Sitong Wei, Qingchao Chen, Dehui Li, Yifan Yang, Yuxin Peng, and Yang Liu. Masked retraining teacher- student framework for domain adaptive object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19039–19049, 2023. 1, 6, 7

    2023

  106. [114]

    End-to-end object detection with adaptive clustering transformer

    Minghang Zheng, Peng Gao, Renrui Zhang, Kunchang Li, Xiaogang Wang, Hongsheng Li, and Hao Dong. End-to-end object detection with adaptive clustering transformer. arXiv preprint arXiv:2011.09315, 2020. 2

    2011 arXiv

  107. [115]

    Ob- jects as points

    Xingyi Zhou, Dequan Wang, and Philipp Krähenbühl. Ob- jects as points. arXiv preprint arXiv:1904.07850, 2019. 2

    1904 arXiv

  108. [116]

    Detecting twenty-thousand classes using image-level supervision

    Xingyi Zhou, Rohit Girdhar, Armand Joulin, Philipp Krähen- bühl, and Ishan Misra. Detecting twenty-thousand classes using image-level supervision. In European Conference on Computer Vision, pages 350–368. Springer, 2022. 2

    2022

  109. [117]

    Adapting object detectors via selective cross-domain alignment

    Xinge Zhu, Jiangmiao Pang, Ceyuan Yang, Jianping Shi, and Dahua Lin. Adapting object detectors via selective cross-domain alignment. In CVPR, pages 687–696, 2019. 3

    2019

  110. [118]

    Deformable detr: Deformable trans- formers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable trans- formers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. 1, 2, 4

    2010 arXiv

  111. [119]

    Object detection in 20 years: A survey

    Zhengxia Zou, Keyan Chen, Zhenwei Shi, Yuhong Guo, and Jieping Ye. Object detection in 20 years: A survey. Proceedings of the IEEE, 111(3):257–276, 2023. 2 TITAN: Query-Token based Domain Adaptive Adversarial Learning Supplementary Material Sec. A provides additional theoretic...

    2023

  112. [120]

    Cityscapes [15] is gathered from urban environments across 50 European cities, provides detailed annotations for 30 semantic classes across 8 categories. It comprises 5,000 high-quality annotated images and a larger set of 20,000 coarsely annotated images, all high-resolution ...

  113. [121]

    Foggy Cityscapes [82] is an extension of the Cityscapes dataset, designed to support research in devel- oping robust computer vision algorithms for autonomous Table 6. Details of Training Datasets Dataset Type Pre-training (Source) Unsupervised Adaptation (Target) Train Val/Te...

  114. [122]

    KITTI [27] is a promenient benchmark for computer vision and robotics, particularly in autonomous driving. Data was collected using a car-mounted sensor suite, including high-resolution color and grayscale cameras, a Velodyne laser scanner, and a GPS/IMU system, in urban, rura...

  115. [123]

    This dataset consists of 10,0000 images of urban environments under different weather conditions, lighting and traffic situations

    SIM10k [46] is a synthetic dataset create using the Grand Theft Auto V (GTA V) engine, it simulates various driving scenarios providing diverse set of high-resolution images with detailed annotations. This dataset consists of 10,0000 images of urban environments under differen...

  116. [124]

    Dataset A to Dataset B

    BDD100k [105] developed by the Berkeley DeepDrive (BDD) team, is one of the largest and most diverse driving video datasets available. It comprises of 100,000 driving videos and 100K keyframe images, and captures a wide range of driving scenarios across urban, suburban and rur...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.