Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

PointOBB-v3: Expanding Performance Boundaries of Single Point-Supervised Oriented Object Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read One point per object is enough for oriented detection, the paper argues.

desk verdict Solid engineering extension of PointOBB with a genuinely new SSFF module and end-to-end branch, but the headline average masks known failure classes and the test-set-tuned hyperparameters weaken the empirical case. read the letter →

arxiv 2501.13898 v2 pith:NZPJUEVJ submitted 2025-01-23 cs.CV cs.AI

Peiyuan Zhang , Junwei Luo , Xue Yang , Yi Yu , Qingyun Li , Yue Zhou , Xiaosong Jia , Xudong Lu
show 4 more authors
Jingdong Chen Xiang Li Junchi Yan Yansheng Li
This is my paper · ORCID
classification cs.CVcs.AI
keywords orientedobjectdetectionsinglepointsupervisionweaklysupervisedlearningremotesensingself-supervisedanglemultipleinstancepseudo-labelgenerationaerialimagery
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

This paper proposes PointOBB-v3, a framework for oriented object detection trained with just one point per object. It claims to generate pseudo rotated boxes without any extra annotations or human priors, and it supports both a two-stage pipeline and an end-to-end variant. The central claim is that combining an original image with a resized view and a rotated/flipped view lets the model learn object scale and orientation from consistency constraints alone: scale-sensitive consistency aligns confidence-score distributions across resolutions, while a self-supervised angle loss exploits object symmetry. The paper reports an average improvement of 3.56 percentage points over previous single point-supervised methods across DIOR-R, DOTA-v1.0/v1.5/v2.0, FAIR1M, STAR, and RSAR, reaching 41.82 mAP on DIOR-R and 50.44 mAP on DOTA-v1.0 with an Oriented R-CNN detector. If true, point labels would become a practical, low-cost annotation for oriented detection in aerial imagery.

What carries the argument

The load-bearing mechanism is the three-view consistency loop. The original view supplies MIL proposal bags; a resized view enforces scale-sensitive consistency of score distributions, fused by the SSFF gating module; and a rotated/flipped view enforces flip and rotation consistency of angle predictions so the angle branch learns symmetry axes despite having no angle labels. The progressive multi-view switching strategy stages these two constraints during training, and the dense-to-sparse matching strategy transfers dense angle estimates to sparse proposals so they can be turned into oriented boxes.

What would settle it

Train and evaluate PointOBB-v3 on a point-supervised test set restricted to bridge, overpass, golf-field, and train-station instances from DIOR-R or DOTA. If the mean angle error on those categories is no better than random while it remains small on symmetric categories, the symmetry assumption is the binding constraint on the claimed gains. A sharper synthetic test is a dataset of identical elongated rectangles with known orientations: if the predicted angles split between two modes for the same shape at 45 degrees, the angle branch cannot disambiguate orientation from symmetry alone.

Watch

Extended reading notes

Core claim

The paper's central claim is that the scale and orientation signals needed for oriented detection can both be recovered from single-point supervision by enforcing two self-consistencies across viewing transformations. Scale is learned by requiring that the MIL confidence-score distribution for the same object from the original and a randomly resized view agree, enforced by the Scale-Sensitive Consistency (SSC) loss, with a gated feature-fusion module making feature-layer selection scale-aware. Orientation is learned by requiring that angle predictions from the original view and a rotated or vertically flipped view obey the known rotation/flip relation, enforced by the Self-Supervised Angle (SSA) loss, which pushes the network to output the object's symmetry axis. A dense-to-sparse matching step attaches these dense angle predictions to the sparse proposals used to form pseudo oriented boxes. Alongside the two-stage pseudo-label pipeline, the paper introduces an end-to-end version that removes the pseudo-label stage, cuts training time by roughly 21.36%, and still outperforms several two-stage predecessors, with an average accuracy gain of 3.56% over prior state-of-the-art methods on seven datasets.

Load-bearing premise

The method's angle supervision assumes objects in aerial images are symmetric about an axis, so flip and rotation consistency can reveal the true orientation; the paper itself concedes that bridges, overpasses, golf fields, and train stations violate this assumption.

Editorial extensions

If this is right

  • A single point label per object can drive an oriented detector to 41.82 mAP on DIOR-R and 50.44 mAP on DOTA-v1.0, cutting annotation cost relative to horizontal or rotated box labels.
  • Pseudo rotated boxes can be generated without human priors such as the sketches used by prior end-to-end point-supervised methods, making the supervision pipeline fully automatic from points.
  • The end-to-end variant removes the pseudo-label stage entirely, reducing training time by roughly 21.36% at a modest accuracy cost, which makes point-supervised oriented detection practical at scale.
  • The same scale and angle consistency constraints transfer across seven aerial benchmarks, including SAR imagery, suggesting the recipe is not tied to one dataset.
  • The gap to fully supervised oriented detection narrows, making point supervision a viable option where oriented box annotation is too expensive.
  • The paper still leaves a performance gap to RBox-supervised detection on categories with extreme aspect ratios or ambiguous boundaries, so future work is motivated on spatial and contextual features.

Reading between the lines

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

  • A testable extension is a multi-hypothesis angle head: near-square symmetric objects have ambiguous symmetry axes, and a single scalar angle may lock onto one mode, causing noise on those instances.
  • The reported average gain is uneven across benchmarks, with large gains on DOTA-v1.0 and smaller gains on FAIR1M and STAR, suggesting the method's benefit scales with category-level symmetry; datasets dominated by elongated or boundary-ambiguous objects may see less.
  • The three-view consistency design could be ported to point-supervised detection in other domains where objects have stable orientations and symmetry axes, such as industrial inspection or microscopy.
  • A sharper stress test is to train the angle branch on identical elongated rectangles with known orientations and check whether predicted angles track the long-axis direction rather than collapsing to one of several symmetric modes.
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. The paper presents PointOBB-v3, a single point-supervised oriented object detection (OOD) framework that extends the authors' prior PointOBB work. The method combines three views (original, resized, rotated/flipped) with a progressive multi-view switching strategy, a Scale-Sensitive Consistency (SSC) loss, a Scale-Sensitive Feature Fusion (SSFF) module, a self-supervised symmetry-based angle acquisition module, and a Dense-to-Sparse matching strategy. It also introduces an end-to-end version with an Instance-Aware Weighting (IAW) strategy. The paper reports state-of-the-art results on DIOR-R, DOTA-v1.0/v1.5/v2.0, FAIR1M, STAR, and RSAR, claiming an average accuracy improvement of 3.56% over prior methods. The authors provide extensive ablations supporting the contribution of each component.

Significance. If the reported results are reliable, this is a meaningful advance for weakly-supervised OOD: a single point per object suffices to train oriented detectors that substantially close the gap to fully supervised performance on standard benchmarks, while an end-to-end variant reduces training time. The paper is thorough in scope, covering seven datasets and including component-level ablations (Tabs. 5, 10) that consistently show gains. The main caveat concerns the generality of the claim: the symmetry-based angle acquisition is shown to fail on elongated or boundary-ambiguous categories, so the headline improvement is an aggregate that may not transfer to such classes.

major comments (4)
  1. [Sec. 4.5, Eq. (12)] The MIL loss in Eq. (12) is written with a second term (1 − Q)(1 − log(1 − S)), which is not the standard binary cross-entropy term (1 − Q) log(1 − S) (with the overall sign arranged appropriately). As written, the loss is mathematically inconsistent with the preceding description and would be a non-standard objective. This is a load-bearing equation for the method, so please correct it and confirm that the implementation matches the intended BCE form.
  2. [Sec. 5.1 and Tabs. 5–10] The ablation studies, including the critical burn-in step selection in Tab. 9 and the grouping/point-range ablations in Tab. 8, are all evaluated on the DIOR-R testing set, and the final results are reported on the same testing set. This creates a risk of test-set overfitting during model selection, especially because the burn-in epochs (6 and 8) were chosen based on these test-set numbers. The authors should either hold out a validation split for these choices or explicitly discuss the implications of using the test set for hyperparameter tuning.
  3. [Sec. 3.2, Sec. 6, Tabs. 1–2] The symmetry-based angle acquisition is the core mechanism for orientation learning, but the paper's own future-work section concedes that it fails for categories with extreme aspect ratios or ambiguous boundaries (Bridge, Overpass, Golf Field, Train Station). The per-category results confirm this: on DIOR-R, PointOBB-v3 (ORCNN) obtains AP50 of 0.1 for Bridge, 0.8 for Overpass, 0.9 for Train Station, and 0.7 for Dam (Tab. 1); on DOTA-v1.0, Bridge is only 23.7 (Tab. 2). The abstract and introduction present PointOBB-v3 as 'a stronger single point-supervised OOD framework' and claim an 'average improvement of 3.56%,' but this aggregate is dominated by symmetric, compact objects and does not hold for the failure classes identified by the authors themselves. The general claim should be qualified to reflect the method's actual limitations, and the failure classes should be mentioned prominently rather than only in a future-work paragraph.
  4. [Sec. 5] All reported mAP values are single-run results without error bars or repeated-seed statistics. This is a concern because several of the claimed improvements are in the 1–3 mAP range (e.g., PointOBB-v3 vs. PointOBB-v2 on DIOR-R is 41.82 vs. 39.62), and the absence of variance estimates makes it difficult to assess whether these gains are significant. The authors should report mean and standard deviation over at least three seeds, or provide a clear justification for why this is not feasible.
minor comments (5)
  1. [Sec. 4.5] In the sentence after Eq. (14), 'α and β keep the same setting with Eq. 14' should refer to Eq. (13), not Eq. (14).
  2. [Sec. 4.2.1] In the text describing the resized view scores, 'Scls_id and Sins_io' appears to be a typo; the second symbol should be 'Sins_id' for consistency with the rest of the notation.
  3. [Sec. 4.3.3, Eq. (9)] Please clarify the angular period convention: if angles are in a π-periodic range, the minimization over k∈Z is appropriate, but the text should state the range explicitly (e.g., [−π/2, π/2)) to avoid ambiguity in the loss definition.
  4. [Sec. 5.2.1] The text 'AP50' is used without a space; ensure consistent formatting throughout.
  5. [Sec. 5.3] Table 4 header contains 'ST AR' with an extra space; minor formatting issue.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: symmetry-based angle recovery is a self-supervised constraint, not a target re-use; failures on BR/OP/GF/TS are a robustness limitation.

full rationale

No circular step was found in PointOBB-v3. The paper is an empirical benchmark study whose central claims are evaluated on held-out test splits of seven external datasets, so the main circularity patterns (fit-then-label-as-prediction, definitional equivalence) do not apply. The angle acquisition module (Sec. 3.2, Eqs. 2-3; Sec. 4.3.3, Eq. 9) enforces flip and rotation equivariance, and the paper shows that for an image symmetric about an axis these constraints force the network output to equal that axis angle; this is a self-supervised consistency derivation from a stated symmetry assumption, not a reuse of the target. The end-to-end branch (Eq. 11) trains the detector head to match the MIL head's own outputs, which is self-training with point-derived weak labels rather than circular benchmarking, since evaluation uses ground-truth OBBs on test sets. Self-citations to PointOBB, PointOBB-v2, and H2RBox-v2 supply the base configuration, baselines, and the symmetry idea, but these are prior published results and the paper's own ablations (Tabs. 5-10) independently support the component choices. The Sec. 6 concession that elongated/ambiguous categories (BR, OP, GF, TS) fail is an assumption-validity and generalization limitation, not a circular reduction; it affects how broadly the 3.56% average improvement transfers, but it does not make the derivation circular. The score of 2, rather than 0, reflects only the presence of minor self-citations that are not load-bearing.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The method is an empirical pipeline; the central claim depends on a set of tuned hyperparameters and domain assumptions rather than on new physical entities. Most free parameters are selected via ablations on the evaluation datasets, which inflates the reported numbers and should ideally be held out. No new particles, forces, or conserved quantities are introduced.

free parameters (9)
  • SSC loss weights omega1, omega2 = 2.0, 1.0
    Weights in Eq. 6, chosen by the authors without reported sensitivity analysis.
  • Stage-switching weights alpha, beta = stage1: alpha=0, beta=1; stages 2/3: alpha=1, beta=0
    Controls which auxiliary view and loss is active during training (Sec 4.5).
  • End-to-end loss weight gamma = 1
    Weight on Le2e in Eq. 15, set without sensitivity study.
  • Finest scale fs = 56
    Used in ROI-to-level mapping and proposal-level assignment (Sec 5.1).
  • Resized view scale factor sigma = random in [0.5, 1.5]
    Range chosen for the resized view (Sec 5.1).
  • Burn-in step1 and step2 epochs = 6 and 8 epochs
    Selected via ablation on the DIOR-R testing set (Tab 9), a case of test-set hyperparameter selection.
  • Point label noise range = 10% of OBB height/width
    Inherited from PointOBB and adopted as default (Sec 5.1); ablations show 10% is better than 0% or 20%.
  • Predefined scale set {s1,...,sG} = not specified
    The SSC loss groups proposals by predefined scales, but the actual scale values are never given (Sec 4.2.1).
  • Number of FPN layers N = 4 (P2 to P5)
    Fusion uses four feature pyramid levels (Sec 4.2.2).
assumptions (4)
  • domain assumption Objects in aerial images have reflection symmetry about their principal orientation axis.
    The SSA loss and flip/rotation consistency in Sec 3.2 and Sec 4.3.3 require symmetric objects; Sec 6 admits this fails for several categories.
  • domain assumption The MIL paradigm's highest-confidence proposal corresponds to the full object extent and scale.
    The scale augmentation module and pseudo-label generation build on the MIL assumption from Sec 3.1 that confident proposals localize the object; the paper's own motivation says this is unstable, hence the SSC loss is added.
  • domain assumption Resized and rotated/flipped views share feature representations and label correspondence with the original view.
    The consistency losses in Sec 4.2 and 4.3 assume exact correspondence between proposals and angle predictions across augmented views.
  • domain assumption ImageNet-pretrained ResNet50/FPN features transfer to aerial imagery.
    All models use ImageNet pretraining (Sec 5.1), standard practice but an external empirical input.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PointOBB-v3: Expanding Performance Boundaries of Single Point-Supervised Oriented Object Detection." pith.science (2026). https://pith.science/paper/NZPJUEVJ

@misc{pith2026250113898,
  author       = {Pith},
  title        = {Pith review of: PointOBB-v3: Expanding Performance Boundaries of Single Point-Supervised Oriented Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NZPJUEVJ}},
  note         = {Machine review of arXiv:2501.13898}
}
read the original abstract

With the growing demand for oriented object detection (OOD), recent studies on point-supervised OOD have attracted significant interest. In this paper, we propose PointOBB-v3, a stronger single point-supervised OOD framework. Compared to existing methods, it generates pseudo rotated boxes without additional priors and incorporates support for the end-to-end paradigm. PointOBB-v3 functions by integrating three unique image views: the original view, a resized view, and a rotated/flipped (rot/flp) view. Based on the views, a scale augmentation module and an angle acquisition module are constructed. In the first module, a Scale-Sensitive Consistency (SSC) loss and a Scale-Sensitive Feature Fusion (SSFF) module are introduced to improve the model's ability to estimate object scale. To achieve precise angle predictions, the second module employs symmetry-based self-supervised learning. Additionally, we introduce an end-to-end version that eliminates the pseudo-label generation process by integrating a detector branch and introduces an Instance-Aware Weighting (IAW) strategy to focus on high-quality predictions. We conducted extensive experiments on the DIOR-R, DOTA-v1.0/v1.5/v2.0, FAIR1M, STAR, and RSAR datasets. Across all these datasets, our method achieves an average improvement in accuracy of 3.56% in comparison to previous state-of-the-art methods. The code will be available at https://github.com/ZpyWHU/PointOBB-v3.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Point2RBox-v2: Rethinking Point-supervised Oriented Object Detection with Spatial Layout Among Instances

    cs.CV 2025-02 conditional novelty 7.0 of 10

    Point2RBox-v2 uses Gaussian overlap, Voronoi watershed, edge, and consistency losses to learn oriented boxes from point annotations, reaching 62.61 AP50 on DOTA-v1.0.

  2. Wholly-WOOD: Wholly Leveraging Diversified-quality Labels for Weakly-supervised Oriented Object Detection

    cs.CV 2025-02 conditional novelty 6.0 of 10

    A unified weakly-supervised framework that, using only horizontal boxes or points, matches or approaches fully RBox-supervised oriented detectors, with point-supervised DOTA-v1.0 AP50 of 62.63, far above prior cited p...

Reference graph

Works this paper leans on

68 extracted references · 60 canonical work pages · cited by 2 Pith papers

  1. [1]

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

    Ding, J., Xue, N., Long, Y., Xia, G.-S., Lu, Q.: Learning roi transformer for oriented object detection in aerial images. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2849–2858 (2019)

    work page 2019

  2. [2]

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

    Han, J., Ding, J., Xue, N., Xia, G.-S.: Redet: A rotation-equivariant detector for aerial object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2786–2795 (2021)

    work page 2021

  3. [3]

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

    Li, Y., Hou, Q., Zheng, Z., Cheng, M.-M., Yang, J., Li, X.: Large selective ker- nel network for remote sensing object detection. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 16794–16805 (2023) 26

    work page 2023

  4. [4]

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

    Xie, X., Cheng, G., Wang, J., Yao, X., Han, J.: Oriented r-cnn for object detection. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 3520–3529 (2021)

    work page 2021

  5. [5]

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

    Yang, X., Yan, J., Feng, Z., He, T.: R3det: Refined single-stage detector with feature refinement for rotating object. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, pp. 3163–3171 (2021)

    work page 2021

  6. [6]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    Li, Y., Luo, J., Zhang, Y., Tan, Y., Yu, J.-G., Bai, S.: Learning to holistically detect bridges from large-size vhr remote sensing imagery. IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    work page 2024

  7. [7]

    In: International Conference on Machine Learning, pp

    Yang, X., Yan, J., Ming, Q., Wang, W., Zhang, X., Tian, Q.: Rethinking rotated object detection with gaussian wasserstein distance loss. In: International Conference on Machine Learning, pp. 11830–11841 (2021). PMLR

    work page 2021

  8. [8]

    Advances in Neural Information Processing Systems 34, 18381–18394 (2021)

    Yang, X., Yang, X., Yang, J., Ming, Q., Wang, W., Tian, Q., Yan, J.: Learn- ing high-precision bounding box for rotated object detection via kullback-leibler divergence. Advances in Neural Information Processing Systems 34, 18381–18394 (2021)

    work page 2021

Show all 68 references
  1. [9]

    arXiv preprint arXiv:2501.09720 (2025)

    Li, Q., Chen, Y., Shu, X., Chen, D., He, X., Yu, Y., Yang, X.: A sim- ple aerial detection baseline of multimodal language models. arXiv preprint arXiv:2501.09720 (2025)

    2025 arXiv

  2. [12]

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

    Feng, X., Yao, X., Cheng, G., Han, J.: Weakly supervised rotation-invariant aerial object detection network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 14146–14155 (2022)

    2022

  3. [15]

    IEEE transactions on pattern analysis and machine intelligence 42(1), 176–191 (2018)

    Tang, P., Wang, X., Bai, S., Shen, W., Bai, X., Liu, W., Yuille, A.: Pcl: Proposal 27 cluster learning for weakly supervised object detection. IEEE transactions on pattern analysis and machine intelligence 42(1), 176–191 (2018)

    2018

  4. [16]

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

    Wan, F., Wei, P., Jiao, J., Han, Z., Ye, Q.: Min-entropy latent model for weakly supervised object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1297–1306 (2018)

    2018

  5. [17]

    In: 2021 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), pp

    Sun, Y., Ran, J., Yang, F., Gao, C., Kurozumi, T., Kimata, H., Ye, Z.: Oriented object detection for remote sensing images based on weakly supervised learn- ing. In: 2021 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), pp. 1–6 (2021). IEEE

    2021

  6. [18]

    In: The Eleventh International Conference on Learning Representations (2023)

    Yang, X., Zhang, G., Li, W., Wang, X., Zhou, Y., Yan, J.: H2rbox: Horizontal box annotation is all you need for oriented object detection. In: The Eleventh International Conference on Learning Representations (2023)

    2023

  7. [19]

    Advances in Neural Information Processing Systems 36 (2024)

    Yu, Y., Yang, X., Li, Q., Zhou, Y., Da, F., Yan, J.: H2rbox-v2: Incorporating sym- metry for boosting horizontal box supervised oriented object detection. Advances in Neural Information Processing Systems 36 (2024)

    2024

  8. [20]

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

    Zhu, T., Ferenczi, B., Purkait, P., Drummond, T., Rezatofighi, H., Van Den Hen- gel, A.: Knowledge combination to learn rotated detection without rotated annotation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 15518–15527 (2023)

    2023

  9. [21]

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

    Cheng, B., Parkhi, O., Kirillov, A.: Pointly-supervised instance segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2617–2626 (2022)

    2022

  10. [22]

    In: European Conference on Computer Vision, pp

    Fan, J., Zhang, Z., Tan, T.: Pointly-supervised panoptic segmentation. In: European Conference on Computer Vision, pp. 319–336 (2022). Springer

    2022

  11. [23]

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

    Li, W., Yuan, Y., Wang, S., Zhu, J., Li, J., Liu, J., Zhang, L.: Point2mask: Point- supervised panoptic segmentation via optimal transport. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 572–581 (2023)

    2023

  12. [25]

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

    Xia, G.-S., Bai, X., Ding, J., Zhu, Z., Belongie, S., Luo, J., Datcu, M., Pelillo, M., Zhang, L.: Dota: A large-scale dataset for object detection in aerial images. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3974–3983 (2018)

    2018

  13. [26]

    In: European Conference on Computer Vision, pp

    Chen, P., Yu, X., Han, X., Hassan, N., Wang, K., Li, J., Zhao, J., Shi, H., Han, Z., Ye, Q.: Point-to-box network for accurate object detection via single point 28 supervision. In: European Conference on Computer Vision, pp. 51–67 (2022). Springer

    2022

  14. [28]

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

    Luo, J., Yang, X., Yu, Y., Li, Q., Yan, J., Li, Y.: Pointobb: Learning ori- ented object detection via single point supervision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 16730–16740 (2024)

    2024

  15. [29]

    arXiv preprint arXiv:2410.08210 (2024)

    Ren, B., Yang, X., Yu, Y., Luo, J., Deng, Z.: Pointobb-v2: Towards simpler, faster, and stronger single point supervised oriented object detection. arXiv preprint arXiv:2410.08210 (2024)

    2024 arXiv

  16. [30]

    IEEE transactions on pattern analysis and machine intelligence 44(4), 1922–1933 (2020)

    Tian, Z., Shen, C., Chen, H., He, T.: Fcos: A simple and strong anchor-free object detector. IEEE transactions on pattern analysis and machine intelligence 44(4), 1922–1933 (2020)

    2020

  17. [31]

    IEEE transactions on pattern analysis and machine intelligence 43(4), 1452–1459 (2020)

    Xu, Y., Fu, M., Wang, Q., Wang, Y., Chen, K., Xia, G.-S., Bai, X.: Glid- ing vertex on the horizontal bounding box for multi-oriented object detection. IEEE transactions on pattern analysis and machine intelligence 43(4), 1452–1459 (2020)

    2020

  18. [32]

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

    Liao, M., Zhu, Z., Shi, B., Xia, G.-s., Bai, X.: Rotation-sensitive regression for ori- ented scene text detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5909–5918 (2018)

    2018

  19. [33]

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

    Liu, X., Liang, D., Yan, S., Chen, D., Qiao, Y., Yan, J.: Fots: Fast oriented text spotting with a unified network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5676–5685 (2018)

    2018

  20. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 45(4), 4335– 4354 (2022)

    Yang, X., Zhang, G., Yang, X., Zhou, Y., Wang, W., Tang, J., He, T., Yan, J.: Detecting rotated objects as gaussian distributions and its 3-d generalization. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(4), 4335– 4354 (2022)

    2022

  21. [35]

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

    Lin, T.-Y., Goyal, P., Girshick, R., He, K., Doll´ ar, P.: Focal loss for dense object detection. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 2980–2988 (2017)

    2017

  22. [36]

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

    Li, W., Chen, Y., Hu, K., Zhu, J.: Oriented reppoints for aerial object detec- tion. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1829–1838 (2022)

    2022

  23. [37]

    IEEE transactions on geoscience and remote sensing 60, 1–11 (2021)

    Han, J., Ding, J., Li, J., Xia, G.-S.: Align deep features for oriented object 29 detection. IEEE transactions on geoscience and remote sensing 60, 1–11 (2021)

    2021

  24. [38]

    In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part VIII 16, pp

    Yang, X., Yan, J.: Arbitrary-oriented object detection with circular smooth label. In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part VIII 16, pp. 677–694 (2020). Springer

    2020

  25. [39]

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

    Yang, X., Hou, L., Zhou, Y., Wang, W., Yan, J.: Dense label encoding for bound- ary discontinuity free rotation detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 15819–15829 (2021)

    2021

  26. [40]

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

    Yu, Y., Da, F.: Phase-shifting coder: Predicting accurate orientation in oriented object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 13354–13363 (2023)

    2023

  27. [41]

    IEEE Transactions on Geoscience and Remote Sensing 61, 1–12 (2023)

    Tan, Z., Jiang, Z., Guo, C., Zhang, H.: Wsodet: A weakly supervised oriented detector for aerial object detection. IEEE Transactions on Geoscience and Remote Sensing 61, 1–12 (2023)

    2023

  28. [42]

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

    Tian, Z., Shen, C., Wang, X., Chen, H.: Boxinst: High-performance instance seg- mentation with box annotations. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5443–5452 (2021)

    2021

  29. [43]

    In: European Conference on Computer Vision, pp

    Li, W., Liu, W., Zhu, J., Cui, M., Hua, X.-S., Zhang, L.: Box-supervised instance segmentation with level set evolution. In: European Conference on Computer Vision, pp. 1–18 (2022). Springer

    2022

  30. [44]

    Neurocomputing 440, 310–320 (2021)

    Iqbal, J., Munir, M.A., Mahmood, A., Ali, A.R., Ali, M.: Leveraging orientation for weakly supervised object detection with application to firearm localization. Neurocomputing 440, 310–320 (2021)

    2021

  31. [45]

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

    Chen, L., Yang, T., Zhang, X., Zhang, W., Sun, J.: Points as queries: Weakly semi- supervised object detection by points. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8823–8832 (2021)

    2021

  32. [46]

    IEEE Transactions on Geoscience and Remote Sensing (2023)

    He, S., Zou, H., Wang, Y., Li, B., Cao, X., Jing, N.: Learning remote sensing object detection with single point supervision. IEEE Transactions on Geoscience and Remote Sensing (2023)

    2023

  33. [47]

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

    Ying, X., Liu, L., Wang, Y., Li, R., Chen, N., Lin, Z., Sheng, W., Zhou, S.: Mapping degeneration meets label evolution: Learning infrared small target detec- tion with single point supervision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,...

    2023

  34. [48]

    arXiv preprint arXiv:2502.04268 (2025) 30

    Yu, Y., Ren, B., Zhang, P., Liu, M., Luo, J., Zhang, S., Da, F., Yan, J., Yang, X.: Point2rbox-v2: Rethinking point-supervised oriented object detection with spatial layout among instances. arXiv preprint arXiv:2502.04268 (2025) 30

    2025 arXiv

  35. [49]

    In: European Conference on Computer Vision, pp

    Bearman, A., Russakovsky, O., Ferrari, V., Fei-Fei, L.: What’s the point: Seman- tic segmentation with point supervision. In: European Conference on Computer Vision, pp. 549–565 (2016). Springer

    2016

  36. [50]

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

    Yu, X., Chen, P., Wu, D., Hassan, N., Li, G., Yan, J., Shi, H., Ye, Q., Han, Z.: Object localization under single coarse point supervision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4868–4877 (2022)

    2022

  37. [51]

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

    Yu, Y., Yang, X., Li, Q., Da, F., Dai, J., Qiao, Y., Yan, J.: Point2rbox: Combine knowledge from synthetic visual patterns for end-to-end oriented object detec- tion with single point supervision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognitio...

    2024

  38. [52]

    Artificial intelligence89(1-2), 31–71 (1997)

    Dietterich, T.G., Lathrop, R.H., Lozano-P´ erez, T.: Solving the multiple instance problem with axis-parallel rectangles. Artificial intelligence89(1-2), 31–71 (1997)

    1997

  39. [53]

    Department of Computer Science & Technology, Nanjing University, Tech

    Zhou, Z.-H.: Multi-instance learning: A survey. Department of Computer Science & Technology, Nanjing University, Tech. Rep 1 (2004)

    2004

  40. [54]

    In: Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Bilen, H., Vedaldi, A.: Weakly supervised deep detection networks. In: Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2846–2854 (2016)

    2016

  41. [55]

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

    Tang, P., Wang, X., Bai, X., Liu, W.: Multiple instance detection network with online instance classifier refinement. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2843–2851 (2017)

    2017

  42. [56]

    IEEE transactions on pattern analysis and machine intelligence 42(1), 176–191 (2018)

    Tang, P., Wang, X., Bai, S., Shen, W., Bai, X., Liu, W., Yuille, A.: Pcl: Proposal cluster learning for weakly supervised object detection. IEEE transactions on pattern analysis and machine intelligence 42(1), 176–191 (2018)

    2018

  43. [57]

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

    Chen, Z., Fu, Z., Jiang, R., Chen, Y., Hua, X.-S.: Slv: Spatial likelihood voting for weakly supervised object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 12995–13004 (2020)

    2020

  44. [58]

    Advances in Neural Information Processing Systems 33, 7005–7019 (2020)

    Shen, Y., Ji, R., Chen, Z., Wu, Y., Huang, F.: Uwsod: Toward fully-supervised- level capacity weakly supervised object detection. Advances in Neural Information Processing Systems 33, 7005–7019 (2020)

    2020

  45. [59]

    IEEE transactions on pattern analysis and machine intelligence 44(9), 5866–5885 (2021)

    Zhang, D., Han, J., Cheng, G., Yang, M.-H.: Weakly supervised object localization and detection: A survey. IEEE transactions on pattern analysis and machine intelligence 44(9), 5866–5885 (2021)

    2021

  46. [60]

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

    Feng, X., Yao, X., Cheng, G., Han, J.: Weakly supervised rotation-invariant aerial object detection network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 14146–14155 (2022) 31

    2022

  47. [61]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 45(10), 11977–11992 (2023)

    Feng, X., Yao, X., Shen, H., Cheng, G., Xiao, B., Han, J.: Learning an invari- ant and equivariant network for weakly supervised object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(10), 11977–11992 (2023)

    2023

  48. [62]

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

    Papadopoulos, D.P., Uijlings, J.R., Keller, F., Ferrari, V.: Training object class detectors with click supervision. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 6374–6383 (2017)

    2017

  49. [63]

    In: European Conference on Computer Vision, pp

    Ren, Z., Yu, Z., Yang, X., Liu, M.-Y., Schwing, A.G., Kautz, J.: Ufo 2: A unified framework towards omni-supervised object detection. In: European Conference on Computer Vision, pp. 288–313 (2020). Springer

    2020

  50. [64]

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

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

    2016

  51. [65]

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

    Lin, T.-Y., Doll´ ar, P., Girshick, R., He, K., Hariharan, B., Belongie, S.: Feature pyramid networks for object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2117–2125 (2017)

    2017

  52. [66]

    IEEE Transactions on Geoscience and Remote Sensing 60, 1–11 (2022)

    Cheng, G., Wang, J., Li, K., Xie, X., Lang, C., Yao, Y., Han, J.: Anchor-free ori- ented proposal generator for object detection. IEEE Transactions on Geoscience and Remote Sensing 60, 1–11 (2022)

    2022

  53. [67]

    ISPRS journal of photogrammetry and remote sensing 159, 296–307 (2020)

    Li, K., Wan, G., Cheng, G., Meng, L., Han, J.: Object detection in opti- cal remote sensing images: A survey and a new benchmark. ISPRS journal of photogrammetry and remote sensing 159, 296–307 (2020)

    2020

  54. [68]

    ISPRS Journal of Photogrammetry and Remote Sensing 184, 116–130 (2022)

    Sun, X., Wang, P., Yan, Z., Xu, F., Wang, R., Diao, W., Chen, J., Li, J., Feng, Y., Xu, T., et al.: Fair1m: A benchmark dataset for fine-grained object recognition in high-resolution remote sensing imagery. ISPRS Journal of Photogrammetry and Remote Sensing 184, 116–130 (2022)

    2022

  55. [69]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    Li, Y., Wang, L., Wang, T., Yang, X., Luo, J., Wang, Q., Deng, Y., Wang, W., Sun, X., Li, H., et al.: Star: A first-ever dataset and a large-scale benchmark for scene graph generation in large-size satellite imagery. IEEE Transactions on Pattern Analysis and Machine Intelligen...

    2024

  56. [70]

    arXiv preprint arXiv:2501.04440 (2025)

    Zhang, X., Yang, X., Li, Y., Yang, J., Cheng, M.-M., Li, X.: Rsar: Restricted state angle resolver and rotated sar benchmark. arXiv preprint arXiv:2501.04440 (2025)

    2025 arXiv

  57. [71]

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

    Zhou, Y., Yang, X., Zhang, G., Wang, J., Liu, Y., Hou, L., Jiang, X., Liu, X., Yan, J., Lyu, C., et al.: Mmrotate: A rotated object detection benchmark using pytorch. In: Proceedings of the 30th ACM International Conference on Multimedia. MM ’22, pp. 7331–7334 (2022) 32

    2022

  58. [72]

    arXiv preprint arXiv:1906.07155 (2019)

    Chen, K., Wang, J., Pang, J., Cao, Y., Xiong, Y., Li, X., Sun, S., Feng, W., Liu, Z., Xu, J., et al.: Mmdetection: Open mmlab detection toolbox and benchmark. arXiv preprint arXiv:1906.07155 (2019)

    2019 arXiv

  59. [73]

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

    Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accurate object detection and semantic segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 580–587 (2014)

    2014

  60. [74]

    In: 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., Fei-Fei, L.: Imagenet: A large- scale hierarchical image database. In: 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp. 248–255 (2009). Ieee 33

    2009

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.