Pith. sign in

REVIEW 4 major objections 6 minor 102 references

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

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Horizontal boxes alone train a detector as accurate as rotated boxes.

desk verdict Solid consolidation with a genuinely new P2R subnet and correct symmetry theory, but the practical label-saving claim rides on idealized derived labels and a thin user study. read the letter →

arxiv 2502.09471 v1 pith:LMJ36KP3 submitted 2025-02-13 cs.CV cs.AI

Yi Yu , Xue Yang , Yansheng Li , Zhenjun Han , Feipeng Da , Junchi Yan This is my paper · ORCID
classification cs.CVcs.AI
keywords orientedobjectdetectionweaklysupervisedlearninghorizontalboundingboxsupervisionpointsymmetry-awareself-supervisedrotationconsistencyrotatedremotesensing
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

Wholly-WOOD is a single detector that can be trained with any mixture of rotated boxes, horizontal boxes, or single points and still output oriented boxes. The paper's central claim is that horizontal-box-only training reaches the accuracy of a fully rotated-box-supervised detector: on five remote sensing benchmarks it averages 0.98% higher AP50 than the rotated-box-trained baseline, and on DOTA-v1.0 it reaches 73.48 AP50 versus 72.44 for the baseline. With point-only supervision it reaches 62.63 AP50 on DOTA-v1.0, far above the previous best at 49.24, at roughly one-quarter of the annotation time. The reason the weaker labels suffice is a symmetry-aware learning principle: a network trained to be consistent under flipping and rotation recovers the reflection axis of a symmetric object, and that axis is the object's orientation.

What carries the argument

The load-bearing identity is Eq. (8): for any reflection-symmetric object $I_{\mathrm{sym}}$ with reflection axis at angle $\theta$, a network $f_{nn}$ that satisfies flip consistency and rotate consistency must output $f_{nn}(I_{\mathrm{sym}}) \equiv \theta \pmod{\pi/2}$. This converts the unlabeled geometry of symmetry into a supervised regression target. Around it, the pipeline is built from the self-supervised branch (two randomly transformed views plus the snap loss), the weakly-supervised branch (Circumscribed-IoU loss against the HBox), and, for point labels, the knowledge-combination module that copies recolored synthetic patterns with known boxes into the image and trains a P2R subnet with a gated fusion-and-scaling mechanism that avoids the FPN anchor-assignment problem for points.

What would settle it

Train the HBox-to-RBox pipeline on a synthetic dataset of asymmetric shapes with known ground-truth angles. If the predicted angles still match the ground truth (rather than failing or collapsing to the HBox-aligned angle), the reflection-symmetry premise is not what carries the method; if they fail specifically for asymmetric shapes, the premise is confirmed as the limiting factor. A cheaper observational check: measure per-category angle error on DOTA and correlate it with human-rated symmetry of the category.

Watch

Extended reading notes

Core claim

At the heart of the method is a mathematical guarantee for symmetry-aware learning. If a visual object is reflection-symmetric about a line at angle $\theta$, and if the network's angle output obeys two self-consistency rules — vertical flipping negates the angle and rotation adds the rotation angle — then the network's prediction is congruent to $\theta$ modulo $\pi/2$. Training the detector with these two consistency losses therefore injects orientation knowledge without any rotated-box label. The horizontal-box annotation supplies position, size, and class through a Circumscribed-IoU loss that compares the predicted box with the horizontal ground truth; the point-annotation branch instead learns box size and angle from synthetic patterns with known boxes that are recolored and pasted around the labeled points. The paper reports that in the HBox-to-RBox setting, Wholly-WOOD matches or slightly beats its own RBox-supervised baseline on DOTA-v1.0/1.5/2.0, HRSC, FAIR1M, and STAR, and that in the Point-to-RBox setting it raises DOTA-v1.0 AP50 by 13.39 points over the previous best point-supervised method.

Load-bearing premise

The method assumes the target objects are reflection-symmetric, and the full angle recovery additionally depends on the weakly-supervised branch resolving the $\theta$ versus $\theta+\pi/2$ ambiguity.

Editorial extensions

If this is right

  • Rotated-box annotation can be replaced by cheaper horizontal-box annotation with no loss in detection accuracy: across DOTA-v1.0/1.5/2.0, HRSC, FAIR1M, and STAR, the HBox-trained model averages 0.98% higher AP50 than the RBox-trained baseline (without multi-scale or random-rotation augmentation).
  • Point-only supervision, which takes about one second per instance to collect, becomes a practical regime: 62.63 AP50 on DOTA-v1.0 versus 49.24 for the previous best point-supervised method, and only 9.81% below the RBox-trained baseline.
  • The framework accepts any mixture of annotation formats in one training run; using 70% points plus 30% HBoxes reaches 72.31 AP50 on DOTA-v1.0, nearly matching the 72.44 of full RBox supervision.
  • The self-supervision branch also helps the fully supervised setting, lifting RBox-trained Wholly-WOOD to 75.63 AP50 on DOTA-v1.0 versus 72.44 for plain rotated FCOS.
  • Training-time RAM drops from 10.10 GB (H2RBox-v2) to 6.67 GB, and inference speed stays at the detector's native rate because the consistency views are used only during training.

Reading between the lines

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

  • The theory pins the angle only up to a quarter turn; the full range is recovered in practice only if the weakly-supervised branch and the object's aspect ratio resolve the $\theta$ versus $\theta+\pi/2$ ambiguity. A testable consequence is that near-square symmetric objects should show concentrated angle errors, and the paper's per-category gaps on Bridge, Soccer-Ball-Field, and Harbor are consist
  • The synthetic-pattern knowledge combination is a general recipe: any label form that fixes a location but not a shape can be paired with recolored category exemplars to synthesize regression targets. That suggests extensions to line segments, scribbles, or partial polygons as supervision.
  • Because the symmetry mechanism uses only consistency of the network output, it should transfer to any domain with symmetric objects, such as PCB inspection, microscopy, or agriculture; the paper shows qualitative results on diatom and PCB images but does not quantify them, so measuring the gap on those domains would be a natural next test.
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 / 6 minor

Summary. The paper proposes Wholly-WOOD, a unified weakly supervised oriented object detection framework that accepts point, horizontal-box (HBox), rotated-box (RBox), and mixed annotations in a single training pipeline. For HBox supervision, it builds on symmetry-aware learning: a self-supervised branch enforces flip and rotate consistency, and a weakly supervised branch uses a Circumscribed IoU loss. For point supervision, it adds a synthetic pattern knowledge-combination module and a novel P2R subnet that fuses FPN layers via gating scores and scales predictions through a continuous binary decoder. The paper gives a theoretical derivation (Sec. III-A) showing that a flip- and rotate-consistent network recovers the reflection angle of a symmetric object up to a quarter turn, and it reports extensive experiments on DOTA-v1.0/1.5/2.0, HRSC, FAIR1M, STAR, and SARDet-100K. The headline results are that HBox-trained Wholly-WOOD achieves 73.48 AP50 on DOTA-v1.0 versus 72.44 for RBox-trained FCOS, and point-trained Wholly-WOOD reaches 62.63 AP50 versus 49.24 for PointOBB-v3. The paper also reports ablations of loss components, view-generation ranges, and annotation noise, and it provides code in PyTorch and Jittor.

Significance. If the results hold, Wholly-WOOD is a significant practical contribution: it provides the first unified treatment of point, HBox, RBox, and mixed annotations for oriented object detection, with a clean theoretical justification for symmetry-aware angle learning and an apparently effective mechanism for point-supervised size and angle regression. The benchmark coverage is broad, the ablations are thorough, and the release of both PyTorch and Jittor implementations is a concrete reproducibility asset. The claimed near-parity with RBox supervision on HBox-to-RBox conversion is empirically well supported on RBox-degraded datasets. The main caveats are that all quantitative results are obtained from RBox ground truths that have been algorithmically degraded to HBoxes or points, and that the real HBox/point-only datasets are evaluated by visualization only; these caveats limit, but do not eliminate, the strength of the practical annotation-saving claim.

major comments (4)
  1. [Sec. III-A, Eq. (8)] The theorem proves only that fnn(Isym) is congruent to the reflection angle theta modulo pi/2, so a flip- and rotate-consistent network cannot by itself distinguish theta from theta+pi/2. The paper states that this is "sufficient for learning rotation in OOD," but it does not explain how the quarter-turn ambiguity is resolved. For HBox supervision, the CircumIoU loss in Sec. III-B can in principle disambiguate because an RBox rotated by pi/2 has a circumscribed HBox with width and height swapped, but this is not stated and is not ablated. For point supervision, the disambiguation would have to come from the synthetic-pattern regression in the P2R subnet, which is likewise not discussed. Please add an explicit statement of the disambiguation mechanism and an ablation that isolates it (for example, a diagnostic on square or near-square objects, or removal of the aspect-ratio cue), since Eq. (8) alone does not justify the sufficiency claim.
  2. [Sec. IV-A and Sec. IV-E] All quantitative AP50 results in Tables VIII and IX are obtained by converting RBox ground truths into minimum-circumscribed HBoxes or center points, as stated in Sec. IV-A. The naturally HBox-annotated dataset SARDet-100K and the diatom and PCB applications are evaluated only by visualization, with Sec. IV-E explicitly noting "there is no ground truth for quantitative analysis." This leaves the abstract's practical claim that HBox/Point annotation can replace RBox annotation at near-par accuracy unverified on real coarse annotations. Human-drawn HBoxes are not guaranteed to be tight circumscribed rectangles, which is the premise of CircumIoU in Sec. III-B, and the noise ablation in Table VI only perturbs height/width multiplicatively without modeling systematic looseness or off-center HBoxes. I recommend evaluating on a held-out RBox-annotated subset of a real HBox-only dataset (e.g., manually annotating RBoxes on a test subset of SARDet-100K or diatom images), or explicitly restricting the headline claim to the derived-label setting.
  3. [Sec. IV-F] The annotation-time reduction figures of 40% (HBox) and 71% (Point), which are reused in the abstract and conclusion, are based on a user study reported only as three average times (1.07 s for Point, 2.23 s for HBox, 3.69 s for RBox). No sample size, number of annotators, annotation protocol, inter-annotator variance, or statistical test is given. Since these percentages are central to the paper's practical motivation, please report the study's methodology and variance, or present the figures as a rough estimate rather than a measured result.
  4. [Sec. IV-C and Table VIII] The abstract and Sec. IV-C describe HBox-trained Wholly-WOOD as performing "very close to that of the RBox-trained counterpart," but the comparison in Sec. IV-C is against the RBox-trained FCOS baseline (72.44 AP50), not against RBox-trained Wholly-WOOD (75.63 AP50 in Table VIII). The gap to the same architecture trained with RBoxes is 2.15 points on DOTA-v1.0 (73.48 vs. 75.63), which is a more direct measure of the annotation-format penalty. Please report both comparisons explicitly, and clarify in the abstract which baseline the phrase "counterpart" refers to.
minor comments (6)
  1. [Sec. III-D, Eq. (27)] The formula for Y uses a single-argument "arctan," but the described continuous binary decoder requires a two-argument atan2. As written, a one-hot G vector with G_1=1 gives Y = N/2 instead of 0 for a single-argument arctan; please use atan2 notation and verify the sign conventions against reference [58].
  2. [Sec. III-D, after Eq. (23)] The sentence "µflp = 1 by default as the weight between rotation and flip has been featured by the proportion of view generation" is confusing. If the view-generation probabilities (95% rotation, 5% flip) functionally replace the explicit weight lambda, please state that clearly and reconcile it with Eq. (23).
  3. [Table VI caption] The noise model is described only in the text; the caption should state the distribution (uniform over (1-sigma, 1+sigma) for HBox width/height and uniform over [-sigma H, +sigma H] for point offsets) and whether the same random seed is used across methods.
  4. [Sec. IV-B] The "Recolor step" ablation is reported in the text (40.27 vs. 28.72 on DOTA) but is not included in any table; adding it to Table VII would make the ablation results easier to compare.
  5. [Fig. 7] The horizontal axis labels such as "99:1," "90:10," etc. are not defined; please clarify whether they denote Point:HBox or Point:RBox ratios, and add an axis label.
  6. [Sec. III-C, Eq. (16)] The variable sigma_r appears in the formula for h but not for w; if this is intentional (e.g., sampling an anisotropic aspect ratio), please say so explicitly, and if it is a typo, correct it.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the angle-recovery theorem is derived in-paper and the headline results are benchmark-supported; flagged gaps are external-validity and limitation issues, not input-output equivalence.

full rationale

No circular step was found. The central theory in Sec. III-A is derived in-paper: Eq. (8), fnn(Isym) ≡ θ (mod π/2), follows directly from the assumed reflection symmetry (Eq. 4) together with the flip and rotate consistency properties (Eqs. 2-3), and the losses in Eq. (9) are optimization targets that encourage those properties rather than relabeled ground-truth angles. The headline HBox- and Point-trained accuracy claims (Tables VIII-IX, Fig. 1c) are evaluated on held-out test splits against independent reference methods, including the RBox-supervised FCOS baseline, BoxInst, SAM-based cascades, the PointOBB series, and other published detectors, so these results are not fitted quantities or renamed inputs. The paper's heavy reliance on the authors' own H2RBox-v2 and Point2RBox is openly disclosed in Sec. I and Sec. II, and the journal version re-derives the components in Secs. III-B and III-C and re-evaluates them in Table VIII, so the self-citations are transparent provenance rather than load-bearing unverified authority. Several limitations are flagged explicitly in the text, and they should be weighed as correctness or external-validity risks rather than circularity: (i) Sec. IV-A states that RBoxes are converted to Points/HBoxes 'by extracting the center point or minimum circumscribed rectangle respectively,' so the quantitative weak-label experiments use idealized coarse labels rather than naturally produced human HBoxes or points; (ii) Sec. IV-E admits for SARDet-100K, diatom, and PCB that 'there is no ground truth for quantitative analysis,' leaving the real-HBox/Point practical claim supported only by visualization; (iii) Sec. III-C discloses a one-shot category exemplar with a known RBox as auxiliary supervision for the Point-to-RBox setting, so the setting is not purely point-only; (iv) Sec. IV-F reports annotation-time savings of 40% and 71% from a user study with no reported sample size or variance; and (v) Eq. (8) guarantees the angle only modulo π/2, and the paper's assertion that this is 'sufficient for learning rotation in OOD' depends on the weakly supervised branch and box aspect ratio to disambiguate the quarter-turn. None of these concerns makes any prediction equal to its input by construction, so the circularity score remains low.

Assumptions & free parameters 6 free parameters · 6 assumptions · 2 invented entities

The framework rests on a small set of explicit assumptions: the two consistency properties, the reflection symmetry of objects, and the implicit disambiguation of the quarter-turn ambiguity by the weakly supervised branch. Most remaining design choices (loss weights, view-generation proportions, pattern opacity, anchor sizes) are ablation-selected constants; they are disclosed in the paper but tuned on the same benchmarks that produce the headline numbers.

free parameters (6)
  • flip-consistency weight lambda = 0.05
    Selected by ablation (Table III) on DOTA and HRSC to maximize AP50; also sets the 95% rotation / 5% flip view-generation proportions in Wholly-WOOD.
  • random rotation range for view generation = pi/4 to 3pi/4
    Selected by ablation (Table IV) on HRSC; ranges near 0 destabilize training, and the range is not derived from theory.
  • Point2RBox view-generation schedule = 66.5% rotation, 3.5% flip, 30% scale
    Ablation-based sampling of transformations for the point-supervised variant (Sec III-C).
  • YOLOF anchor size for point supervision = 64x64 (DOTA), 128x128 (others)
    Hand-set in Point2RBox (Sec III-C) to compensate for the missing size information in point labels.
  • pattern blending parameters = alpha0=0.1, alpha1=0.9, k0,k1 uniform in [0.1,2], sigma_base ~ N(0,0.4)
    Hand-chosen constants in the synthetic pattern generator (Eqs. 16 and 17); the recolor step itself is ablated (40.27 vs 28.72 without it).
  • label assignment thresholds = L1 distance threshold 32; K=4 nearest anchors; NMS IoU 0.05
    Constants in Eq. (18) and Sec III-C that control which anchors and which synthetic patterns participate in training.
assumptions (6)
  • domain assumption Flip consistency (Property I, Eq. 2): the network output changes sign under vertical flip of the input.
    Assumed achievable by training with Lflp (Eq. 9); Table I shows convergence requires the PSC coder and snap loss, so the property is not free.
  • domain assumption Rotate consistency (Property II, Eq. 3): the output rotates by R when the input is rotated by R.
    Assumed achievable by training with Lrot (Eq. 9); requires cross-view object matching by the FCOS assigner (Sec III-B).
  • domain assumption Target objects possess reflection symmetry about their principal axis (Eq. 4).
    This is the load-bearing premise; it is violated by several DOTA categories (Bridge, Harbor, Soccer-Ball-Field) and Fig. 1b shows larger gaps there.
  • standard math A reflection is the composition of a rotation by 2*theta and a vertical flip (Eq. 6).
    Elementary linear algebra; verified by direct matrix multiplication.
  • domain assumption Oriented boxes are defined modulo pi, so the snap loss (Eq. 10) can target any k*pi + theta_target.
    Standard long-side convention in oriented object detection; the PSC coder and snap loss are needed for stable training (Table I).
  • ad hoc to paper The weakly-supervised branch resolves the pi/2 ambiguity left by Eq. (8).
    Not stated in the theory; the CircumIoU loss with the HBox extent and object aspect ratio must select between theta and theta+pi/2. Without this, the theory only guarantees the angle modulo pi/2.
invented entities (2)
  • Synthetic pattern knowledge combination (per-category one-shot crop, recolored and blended into training images) independent evidence
    purpose: Provides known-RBox regression targets for point-annotated objects, since points carry no size or angle information (Sec III-C, Eqs. 14 to 17).
    The Point-to-RBox benchmark results (62.63 AP50 on DOTA vs 40.27 for the no-fusion Point2RBox) and the recolor ablation (40.27 vs 28.72) give a falsifiable handle on the held-out test set.
  • P2R subnet gating score G_n and scale factor m = 2^Y (Eqs. 25 to 28) independent evidence
    purpose: Anchor-free multi-scale box regression for point-supervised objects: fuses FPN layers by learned gates and rescales predicted boxes.
    Ablation Table VII (+9.61 AP50 over no-fusion Wholly-WOOD) and Fig. 6 show the mechanism changes performance, and the learned scale factor differs observably for small, median, and large objects.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Wholly-WOOD: Wholly Leveraging Diversified-quality Labels for Weakly-supervised Oriented Object Detection." pith.science (2026). https://pith.science/paper/LMJ36KP3

@misc{pith2026250209471,
  author       = {Pith},
  title        = {Pith review of: Wholly-WOOD: Wholly Leveraging Diversified-quality Labels for Weakly-supervised Oriented Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LMJ36KP3}},
  note         = {Machine review of arXiv:2502.09471}
}
read the original abstract

Accurately estimating the orientation of visual objects with compact rotated bounding boxes (RBoxes) has become a prominent demand, which challenges existing object detection paradigms that only use horizontal bounding boxes (HBoxes). To equip the detectors with orientation awareness, supervised regression/classification modules have been introduced at the high cost of rotation annotation. Meanwhile, some existing datasets with oriented objects are already annotated with horizontal boxes or even single points. It becomes attractive yet remains open for effectively utilizing weaker single point and horizontal annotations to train an oriented object detector (OOD). We develop Wholly-WOOD, a weakly-supervised OOD framework, capable of wholly leveraging various labeling forms (Points, HBoxes, RBoxes, and their combination) in a unified fashion. By only using HBox for training, our Wholly-WOOD achieves performance very close to that of the RBox-trained counterpart on remote sensing and other areas, significantly reducing the tedious efforts on labor-intensive annotation for oriented objects. The source codes are available at https://github.com/VisionXLab/whollywood (PyTorch-based) and https://github.com/VisionXLab/whollywood-jittor (Jittor-based).

Figures

Figures reproduced from arXiv: 2502.09471 by the authors.

Figure 1
Figure 1. To illustrate the task we aim at and the results we achieve. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overview of H2RBox-v2. (a) Self-supervised (SS) branch that learns the orientation from the symmetry of objects. (b) Weakly-supervised branch that learns other properties from HBoxes. (c) Snap loss for the SS branch. (d) Circumscribed IoU (CircumIoU) loss for the WS branch. Table I for ablation). This equation can be done by using the modulo operation in the code implementation. Finally, the loss for the SS bran… view at source ↗
Figure 3
Figure 3. The training flowchart of Point2RBox, consisting of knowledge combination and transform self-supervision. The core idea is to combine knowledge [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The Wholly-WOOD architecture. It features two key components: [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The architecture of P2R subnet. The multiple output layers of the FPN automatically aggregate based on the gating score activated by each layer [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: An experiment to show that the P2R subnet can scale the output boxes [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Experiments with different Point/HBox/RBox proportions on the [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Experimental results of our Wholly-WOOD. From left to right: [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Visualization to compare the state-of-the-art approaches and our Wholly-WOOD. The first three columns are HBox-to-RBox methods and the others [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

102 extracted references · 80 canonical work pages

  1. [1]

    Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks,

    X. Yang, H. Sun, K. Fu, J. Yang, X. Sun, M. Yan, and Z. Guo, “Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks,” Remote Sensing, vol. 10, no. 1, 2018. 1

    2018

  2. [2]

    Dota: A large- scale dataset for object detection in aerial images,

    G.-S. Xia, X. Bai, J. Ding, Z. Zhu, S. Belongie, J. Luo, M. Datcu, M. Pelillo, and L. Zhang, “Dota: A large- scale dataset for object detection in aerial images,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 3974–3983. 1, 8, 14

    2018

  3. [3]

    Remote sensing object detection in the deep learning era—a review,

    S. Gui, S. Song, R. Qin, and Y . Tang, “Remote sensing object detection in the deep learning era—a review,” Remote Sensing, vol. 16, no. 2, 2024. 1

    2024

  4. [4]

    A high res- olution optical satellite image dataset for ship recogni- tion and some new baselines,

    Z. Liu, L. Yuan, L. Weng, and Y . Yang, “A high res- olution optical satellite image dataset for ship recogni- tion and some new baselines,” in International Confer- ence on Pattern Recognition Applications and Methods, vol. 2, 2017, pp. 324–331. 1, 8

    2017

  5. [5]

    Diffdet4sar: Diffusion-based aircraft target detection network for sar images,

    J. Zhou, C. Xiao, B. Peng, Z. Liu, L. Liu, Y . Liu, and X. Li, “Diffdet4sar: Diffusion-based aircraft target detection network for sar images,” IEEE Geoscience and Remote Sensing Letters , 2024. 1

    2024

  6. [6]

    Deep learning-based binocular image analysis for in situ measurement of particle length distribution during crystallization process,

    J. Fan, T. Liu, Y . Shuang, B. Song, J. Chen, and Y . Tan, “Deep learning-based binocular image analysis for in situ measurement of particle length distribution during crystallization process,” IEEE Transactions on Instrumentation and Measurement , 2023. 1

    2023

  7. [7]

    Rotatedstom- atanet: a deep rotated object detection network for direc- tional stomata phenotype analysis,

    X. Yang, J. Wang, F. Li, C. Zhou, M. Wu, C. Zheng, L. Yang, Z. Li, Y . Li, S. Guo et al. , “Rotatedstom- atanet: a deep rotated object detection network for direc- tional stomata phenotype analysis,” Plant Cell Reports , vol. 43, no. 5, pp. 1–18, 2024. 1

    2024

  8. [8]

    An oriented object detector towards diatoms,

    S. Gong, K. Wu, Z. Xia, L. Ran, C. Gu, C. Lu, T. Guan, and Y . Zhao, “An oriented object detector towards diatoms,” in International Joint Conference on Neural Networks. IEEE, 2023, pp. 1–8. 1

    2023

Show all 102 references
  1. [9]

    Deep multi-modal object detection and semantic seg- mentation for autonomous driving: Datasets, meth- ods, and challenges,

    D. Feng, C. Haase-Sch ¨utz, L. Rosenbaum, H. Hertlein, C. Gl ¨aser, F. Timm, W. Wiesbeck, and K. Dietmayer, “Deep multi-modal object detection and semantic seg- mentation for autonomous driving: Datasets, meth- ods, and challenges,” IEEE Transactions on Intelligent JOURNAL OF ...

    2021

  2. [10]

    Grasp pose detection from a single rgb image,

    H. Cheng, Y . Wang, and M. Q.-H. Meng, “Grasp pose detection from a single rgb image,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems, 2021, pp. 4686–4691. 1

    2021

  3. [11]

    Exploring rotated object detection models for antipodal robotic grasping,

    V . Holomjova and P. Meißner, “Exploring rotated object detection models for antipodal robotic grasping,” in UK Robotics and Autonomous Systems Conference, 2022. 1

    2022

  4. [12]

    Detecting lesion bounding ellipses with gaussian proposal networks,

    Y . Li, “Detecting lesion bounding ellipses with gaussian proposal networks,” in Machine Learning in Medical Imaging: 10th International Workshop , 2019, pp. 337–

    2019

  5. [13]

    Arbitrary-oriented scene text detection via rotation proposals,

    J. Ma, W. Shao, H. Ye, L. Wang, H. Wang, Y . Zheng, and X. Xue, “Arbitrary-oriented scene text detection via rotation proposals,” IEEE Transactions on Multimedia , vol. 20, no. 11, pp. 3111–3122, 2018. 1

    2018

  6. [14]

    Rotation-sensitive regression for oriented scene text detection,

    M. Liao, Z. Zhu, B. Shi, G.-S. Xia, and X. Bai, “Rotation-sensitive regression for oriented scene text detection,” in IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition , 2018, pp. 5909–5918. 1

    2018

  7. [15]

    East: An efficient and accurate scene text detector,

    X. Zhou, C. Yao, H. Wen, Y . Wang, S. Zhou, W. He, and J. Liang, “East: An efficient and accurate scene text detector,” in IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2642–2651. 1

    2017

  8. [16]

    Dynamic refinement network for oriented and densely packed object detection,

    X. Pan, Y . Ren, K. Sheng, W. Dong, H. Yuan, X. Guo, C. Ma, and C. Xu, “Dynamic refinement network for oriented and densely packed object detection,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 11 207–11 216. 1

    2020

  9. [17]

    Pcbssd: Self-supervised symmetry-aware detector for pcb displacement and ori- entation inspection,

    J. Li, F. Da, and Y . Yu, “Pcbssd: Self-supervised symmetry-aware detector for pcb displacement and ori- entation inspection,” Measurement, vol. 243, p. 116342,

  10. [18]

    Wsrd-net: A convolutional neural network-based arbitrary-oriented wheat stripe rust detection method,

    H. Liu, L. Jiao, R. Wang, C. Xie, J. Du, H. Chen, and R. Li, “Wsrd-net: A convolutional neural network-based arbitrary-oriented wheat stripe rust detection method,” Frontiers in Plant Science , vol. 13, p. 876069, 2022. 1

    2022

  11. [19]

    A deep learning method for oriented and small wheat spike detection (oswsdet) in uav images,

    J. Zhao, J. Yan, T. Xue, S. Wang, X. Qiu, X. Yao, Y . Tian, Y . Zhu, W. Cao, and X. Zhang, “A deep learning method for oriented and small wheat spike detection (oswsdet) in uav images,” Computers and Electronics in Agriculture, vol. 198, p. 107087, 2022. 1

    2022

  12. [20]

    Precise maize detasseling base on oriented object detection for tas- sels,

    C. Song, F. Zhang, J. Li, and J. Zhang, “Precise maize detasseling base on oriented object detection for tas- sels,” Computers and Electronics in Agriculture , vol. 202, p. 107382, 2022. 1

    2022

  13. [21]

    Real- time rotation-invariant face detection with progressive calibration networks,

    X. Shi, S. Shan, M. Kan, S. Wu, and X. Chen, “Real- time rotation-invariant face detection with progressive calibration networks,” in IEEE Conference on Computer Vision and Pattern Recognition , 2018, pp. 2295–2303. 1

    2018

  14. [22]

    Detecting insulator strings as linked chain structure in smart grid inspection,

    N. Wei, X. Li, J. Jin, P. Chen, and S. Sun, “Detecting insulator strings as linked chain structure in smart grid inspection,” IEEE Transactions on Industrial Informat- ics, 2022. 1

    2022

  15. [23]

    Automatic overheating defect diagnosis based on rotated detector for insulator in infrared image,

    Y . Lin, L. Tian, and Q. Du, “Automatic overheating defect diagnosis based on rotated detector for insulator in infrared image,” IEEE Sensors Journal , 2023. 1

    2023

  16. [24]

    Yudo: Yolo for uniform directed object detection,

    D. Nedeljkovic, “Yudo: Yolo for uniform directed object detection,” arXiv preprint arXiv:2308.04542 , 2023. 1

    2023 arXiv

  17. [25]

    Aitars-net: A novel network for detecting arbitrary-oriented transverse aeolian ridges from tianwen-1 hiric images,

    Z. Cao, Z. Kang, T. Hu, Z. Yang, D. Chen, X. Ren, Q. Meng, and D. Wang, “Aitars-net: A novel network for detecting arbitrary-oriented transverse aeolian ridges from tianwen-1 hiric images,” ISPRS Journal of Pho- togrammetry and Remote Sensing , vol. 211, pp. 135– 155, 2024. 1

    2024

  18. [26]

    Deep learning for generic object detection: A survey,

    L. Liu, W. Ouyang, X. Wang, P. Fieguth, J. Chen, X. Liu, and M. Pietik ¨ainen, “Deep learning for generic object detection: A survey,” International Journal of Computer Vision, vol. 128, no. 2, pp. 261–318, 2020. 1, 2

    2020

  19. [27]

    Star: A first-ever dataset and a large- scale benchmark for scene graph generation in large- size satellite imagery,

    Y . Li, L. Wang, T. Wang, X. Yang, J. Luo, Q. Wang, Y . Deng, W. Wang, X. Sun, H. Li, B. Dang, Y . Zhang, Y . Yu, and J. Yan, “Star: A first-ever dataset and a large- scale benchmark for scene graph generation in large- size satellite imagery,” IEEE Transactions on Pattern Ana...

    2024

  20. [28]

    On the arbitrary-oriented object detection: Classification based approaches revisited,

    X. Yang and J. Yan, “On the arbitrary-oriented object detection: Classification based approaches revisited,” International Journal of Computer Vision , vol. 130, pp. 1340–1365, 2022. 1

    2022

  21. [29]

    Detecting rotated objects as gaussian distributions and its 3-d generalization,

    X. Yang, G. Zhang, X. Yang, Y . Zhou, W. Wang, J. Tang, T. He, and J. Yan, “Detecting rotated objects as gaussian distributions and its 3-d generalization,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 4, pp. 4335–4354, 2023. 1, 3

    2023

  22. [30]

    Remote sensing object detection meets deep learning: A metareview of challenges and ad- vances,

    X. Zhang, T. Zhang, G. Wang, P. Zhu, X. Tang, X. Jia, and L. Jiao, “Remote sensing object detection meets deep learning: A metareview of challenges and ad- vances,” IEEE Geoscience and Remote Sensing Mag- azine, vol. 11, no. 4, pp. 8–44, 2023. 1

    2023

  23. [31]

    Ef- ficient inductive vision transformer for oriented object detection in remote sensing imagery,

    C. Zhang, J. Su, Y . Ju, K.-M. Lam, and Q. Wang, “Ef- ficient inductive vision transformer for oriented object detection in remote sensing imagery,” IEEE Transac- tions on Geoscience and Remote Sensing , vol. 61, pp. 1–20, 2023. 1

    2023

  24. [32]

    A compre- hensive survey of oriented object detection in remote sensing images,

    L. Wen, Y . Cheng, Y . Fang, and X. Li, “A compre- hensive survey of oriented object detection in remote sensing images,” Expert Systems with Applications , p. 119960, 2023. 1, 2

    2023

  25. [33]

    Theoretically achieving continuous representation of oriented bounding boxes,

    Z. Xiao, G. Yang, X. Yang, T. Mu, J. Yan, and S. Hu, “Theoretically achieving continuous representation of oriented bounding boxes,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 16 912–16 922. 1

    2024

  26. [34]

    Rotation-aware and multi-scale convolutional neural network for object detection in remote sensing images,

    K. Fu, Z. Chang, Y . Zhang, G. Xu, K. Zhang, and X. Sun, “Rotation-aware and multi-scale convolutional neural network for object detection in remote sensing images,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 161, pp. 294–308, 2020. 1

    2020

  27. [35]

    Learn- ing roi transformer for oriented object detection in aerial images,

    J. Ding, N. Xue, Y . Long, G.-S. Xia, and Q. Lu, “Learn- ing roi transformer for oriented object detection in aerial images,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 2849–2858. 1, 3 JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 16

    2019

  28. [36]

    Oriented r-cnn for object detection,

    X. Xie, G. Cheng, J. Wang, X. Yao, and J. Han, “Oriented r-cnn for object detection,” in IEEE/CVF International Conference on Computer Vision, 2021, pp. 3520–3529. 1, 3

    2021

  29. [37]

    Redet: A rotation-equivariant detector for aerial object detection,

    J. Han, J. Ding, N. Xue, and G.-S. Xia, “Redet: A rotation-equivariant detector for aerial object detection,” in IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, 2021, pp. 2785–2794. 1, 3, 10

    2021

  30. [38]

    What’s the point: Semantic segmentation with point supervision,

    A. Bearman, O. Russakovsky, V . Ferrari, and L. Fei-Fei, “What’s the point: Semantic segmentation with point supervision,” in European Conference on Computer Vision, 2016, pp. 549–565. 1

    2016

  31. [39]

    Object detection in optical remote sensing images: A survey and a new benchmark,

    K. Li, G. Wan, G. Cheng, L. Meng, and J. Han, “Object detection in optical remote sensing images: A survey and a new benchmark,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 159, pp. 296–307, 2020. 1

    2020

  32. [40]

    Sardet-100k: Towards open-source benchmark and toolkit for large-scale sar object detection,

    Y . Li, X. Li, W. Li, Q. Hou, L. Liu, M.-M. Cheng, and J. Yang, “Sardet-100k: Towards open-source benchmark and toolkit for large-scale sar object detection,” in Advances in Neural Information Processing Systems ,

  33. [41]

    Anchor-free oriented proposal generator for object detection,

    G. Cheng, J. Wang, K. Li, X. Xie, C. Lang, Y . Yao, and J. Han, “Anchor-free oriented proposal generator for object detection,” IEEE Transactions on Geoscience and Remote Sensing , 2022. 1

    2022

  34. [42]

    Fcos: Fully convolutional one-stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: Fully convolutional one-stage object detection,” in IEEE/CVF International Conference on Computer Vision, 2019, pp. 9626–9635. 2, 3, 6, 7, 9, 10, 11, 12

    2019

  35. [43]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.- Y . Lo, P. Dollar, and R. Girshick, “Segment anything,” in IEEE/CVF International Conference on Computer Vision, 2023, pp. 4015–4026. 2, 3, 11, 12

    2023

  36. [44]

    2017 iccv chal- lenge: Detecting symmetry in the wild,

    C. Funk, S. Lee, M. R. Oswald, S. Tsogkas, W. Shen, A. Cohen, S. Dickinson, and Y . Liu, “2017 iccv chal- lenge: Detecting symmetry in the wild,” in IEEE In- ternational Conference on Computer Vision Workshops, 2017, pp. 1692–1701. 2

    2017

  37. [45]

    H2rbox-v2: Incorporating symmetry for boosting hor- izontal box supervised oriented object detection,

    Y . Yu, X. Yang, Q. Li, Y . Zhou, F. Da, and J. Yan, “H2rbox-v2: Incorporating symmetry for boosting hor- izontal box supervised oriented object detection,” in Advances in Neural Information Processing Systems ,

  38. [46]

    Point2rbox: Combine knowledge from synthetic vi- sual patterns for end-to-end oriented object detection with single point supervision,

    Y . Yu, X. Yang, Q. Li, F. Da, J. Dai, Y . Qiao, and J. Yan, “Point2rbox: Combine knowledge from synthetic vi- sual patterns for end-to-end oriented object detection with single point supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogn...

    2024

  39. [47]

    Pytorch: An imperative style, high- performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “Pytorch: An imperative style, high- p...

    2019

  40. [48]

    Jittor: a novel deep learning framework with meta-operators and unified graph execution,

    S.-M. Hu, D. Liang, G.-Y . Yang, G.-W. Yang, and W.- Y . Zhou, “Jittor: a novel deep learning framework with meta-operators and unified graph execution,” Science China Information Sciences , vol. 63, no. 222103, pp. 1–21, 2020. 2

    2020

  41. [49]

    Object detection with deep learning: A review,

    Z.-Q. Zhao, P. Zheng, S.-T. Xu, and X. Wu, “Object detection with deep learning: A review,” IEEE Trans- actions on Neural Networks and Learning Systems , vol. 30, no. 11, pp. 3212–3232, 2019. 2

    2019

  42. [50]

    Zone evaluation: Revealing spatial bias in object detection,

    Z. Zheng, Y . Chen, Q. Hou, X. Li, P. Wang, and M.- M. Cheng, “Zone evaluation: Revealing spatial bias in object detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence , pp. 1–16, 2024. 2

    2024

  43. [51]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” IEEE Trans- actions on Pattern Analysis and Machine Intelligence , vol. 42, no. 2, pp. 318–327, 2020. 3, 11, 12

    2020

  44. [52]

    R3det: Refined single-stage detector with feature refinement for rotating object,

    X. Yang, J. Yan, Z. Feng, and T. He, “R3det: Refined single-stage detector with feature refinement for rotating object,” in AAAI Conference on Artificial Intelligence , vol. 35, 2021, pp. 3163–3171. 3, 11

    2021

  45. [53]

    Align deep fea- tures for oriented object detection,

    J. Han, J. Ding, J. Li, and G.-S. Xia, “Align deep fea- tures for oriented object detection,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–11,

  46. [54]

    Scrdet: Towards more robust detection for small, cluttered and rotated objects,

    X. Yang, J. Yang, J. Yan, Y . Zhang, T. Zhang, Z. Guo, X. Sun, and K. Fu, “Scrdet: Towards more robust detection for small, cluttered and rotated objects,” in IEEE/CVF International Conference on Computer Vi- sion, 2019, pp. 8231–8240. 3

    2019

  47. [55]

    Learning modulated loss for rotated object detection,

    W. Qian, X. Yang, S. Peng, J. Yan, and Y . Guo, “Learning modulated loss for rotated object detection,” in AAAI Conference on Artificial Intelligence , vol. 35, 2021, pp. 2458–2466. 3

    2021

  48. [56]

    Arbitrary-oriented object detection with circular smooth label,

    X. Yang and J. Yan, “Arbitrary-oriented object detection with circular smooth label,” in European Conference on Computer Vision, 2020, pp. 677–694. 3

    2020

  49. [57]

    Dense label encoding for boundary discontinuity free rotation detection,

    X. Yang, L. Hou, Y . Zhou, W. Wang, and J. Yan, “Dense label encoding for boundary discontinuity free rotation detection,” in IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, 2021, pp. 15 814–15 824. 3

    2021

  50. [58]

    On boundary discontinuity in angle regression based arbitrary oriented object detection,

    Y . Yu and F. Da, “On boundary discontinuity in angle regression based arbitrary oriented object detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1–14, 2024. 3, 8

    2024

  51. [59]

    Rethinking rotated object detection with gaus- sian wasserstein distance loss,

    X. Yang, J. Yan, M. Qi, W. Wang, X. Zhang, and T. Qi, “Rethinking rotated object detection with gaus- sian wasserstein distance loss,” in 38th International Conference on Machine Learning , vol. 139, 2021, pp. 11 830–11 841. 3, 11, 12

    2021

  52. [60]

    Learning high-precision bounding box for rotated object detection via kullback-leibler diver- gence,

    X. Yang, X. Yang, J. Yang, Q. Ming, W. Wang, Q. Tian, and J. Yan, “Learning high-precision bounding box for rotated object detection via kullback-leibler diver- gence,” in Advances in Neural Information Processing Systems, vol. 34, 2021, pp. 18 381–18 394. 3, 11 JOURNAL OF LAT...

    2021

  53. [61]

    The kfiou loss for rotated object detection,

    X. Yang, Y . Zhou, G. Zhang, J. Yang, W. Wang, J. Yan, X. Zhang, and Q. Tian, “The kfiou loss for rotated object detection,” in International Conference on Learning Representations , 2023. 3, 11

    2023

  54. [62]

    Rep- points: Point set representation for object detection,

    Z. Yang, S. Liu, H. Hu, L. Wang, and S. Lin, “Rep- points: Point set representation for object detection,” in IEEE/CVF International Conference on Computer Vision, 2019, pp. 9656–9665. 3, 11

    2019

  55. [63]

    G-rep: Gaussian representation for arbitrary-oriented object detection,

    L. Hou, K. Lu, X. Yang, Y . Li, and J. Xue, “G-rep: Gaussian representation for arbitrary-oriented object detection,” Remote Sensing, vol. 15, no. 3, 2023. 3

    2023

  56. [64]

    Oriented reppoints for aerial object detection,

    W. Li, Y . Chen, K. Hu, and J. Zhu, “Oriented reppoints for aerial object detection,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 1829–1838. 3, 11

    2022

  57. [65]

    A simple aerial detection base- line of multimodal language models,

    Q. Li, Y . Chen, X. Shu, D. Chen, X. He, Y . Yu, and X. Yang, “A simple aerial detection base- line of multimodal language models,” arXiv preprint arXiv:2501.09720, 2025. 3

    2025 arXiv

  58. [66]

    Leveraging orientation for weakly supervised object detection with application to firearm localiza- tion,

    J. Iqbal, M. A. Munir, A. Mahmood, A. R. Ali, and M. Ali, “Leveraging orientation for weakly supervised object detection with application to firearm localiza- tion,” Neurocomputing, vol. 440, pp. 310–320, 2021. 3

    2021

  59. [67]

    Oriented object detection for remote sensing images based on weakly supervised learning,

    Y . Sun, J. Ran, F. Yang, C. Gao, T. Kurozumi, H. Ki- mata, and Z. Ye, “Oriented object detection for remote sensing images based on weakly supervised learning,” in IEEE International Conference on Multimedia & Expo Workshops, 2021, pp. 1–6. 3, 12

    2021

  60. [68]

    Knowledge combination to learn rotated detection without rotated annotation,

    T. Zhu, B. Ferenczi, P. Purkait, T. Drummond, H. Rezatofighi, and A. van den Hengel, “Knowledge combination to learn rotated detection without rotated annotation,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023. 3, 12, 13

    2023

  61. [69]

    Simple does it: Weakly supervised instance and semantic segmentation,

    A. Khoreva, R. Benenson, J. Hosang, M. Hein, and B. Schiele, “Simple does it: Weakly supervised instance and semantic segmentation,” in IEEE Conference on Computer Vision and Pattern Recognition , 2017, pp. 876–885. 3

    2017

  62. [70]

    Weakly supervised instance segmentation using the bounding box tightness prior,

    C.-C. Hsu, K.-J. Hsu, C.-C. Tsai, Y .-Y . Lin, and Y .- Y . Chuang, “Weakly supervised instance segmentation using the bounding box tightness prior,” Advances in Neural Information Processing Systems , vol. 32, 2019. 3

    2019

  63. [71]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in IEEE International Conference on Computer Vision, 2017, pp. 2961–2969. 3

    2017

  64. [72]

    Box- inst: High-performance instance segmentation with box annotations,

    Z. Tian, C. Shen, X. Wang, and H. Chen, “Box- inst: High-performance instance segmentation with box annotations,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 5443–5452. 3, 11

    2021

  65. [73]

    Conditional convolu- tions for instance segmentation,

    Z. Tian, C. Shen, and H. Chen, “Conditional convolu- tions for instance segmentation,” in European Confer- ence on Computer Vision , 2020, pp. 282–298. 3

    2020

  66. [74]

    Box-supervised instance segmentation with level set evolution,

    W. Li, W. Liu, J. Zhu, M. Cui, X. Hua, and L. Zhang, “Box-supervised instance segmentation with level set evolution,” in European Conference on Computer Vi- sion, 2022. 3, 11, 12

    2022

  67. [75]

    H2rbox: Horizontal box annotation is all you need for oriented object detection,

    X. Yang, G. Zhang, W. Li, X. Wang, Y . Zhou, and J. Yan, “H2rbox: Horizontal box annotation is all you need for oriented object detection,” International Con- ference on Learning Representations , 2023. 3, 9, 11, 12, 13

    2023

  68. [76]

    Explicit and implicit box equivariance learning for weakly-supervised rotated object detection,

    L. Wang, Y . Zhan, X. Lin, B. Yu, L. Ding, J. Zhu, and D. Tao, “Explicit and implicit box equivariance learning for weakly-supervised rotated object detection,” IEEE Transactions on Emerging Topics in Computational Intelligence, 2024. 3, 11

    2024

  69. [77]

    Afws: Angle-free weakly-supervised rotating object detection for remote sensing images,

    J. Lu, Q. Hu, R. Zhu, Y . Wei, and T. Li, “Afws: Angle-free weakly-supervised rotating object detection for remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing , 2024. 3, 11, 12

    2024

  70. [78]

    Point-to-box network for accurate object detection via single point supervision,

    P. Chen, X. Yu, X. Han, N. Hassan, K. Wang, J. Li, J. Zhao, H. Shi, Z. Han, and Q. Ye, “Point-to-box network for accurate object detection via single point supervision,” in European Conference on Computer Vision, 2022. 3, 11, 12

    2022

  71. [79]

    Weakly-supervised salient object de- tection using point supervision,

    S. Gao, W. Zhang, Y . Wang, Q. Guo, C. Zhang, Y . He, and W. Zhang, “Weakly-supervised salient object de- tection using point supervision,” in AAAI Conference on Artificial Intelligence, vol. 36, 2022, pp. 670–678. 3

    2022

  72. [80]

    Mapping degeneration meets label evolution: Learning infrared small target detection with single point supervision,

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

    2023

  73. [81]

    Points as queries: Weakly semi-supervised object de- tection by points,

    L. Chen, T. Yang, X. Zhang, W. Zhang, and J. Sun, “Points as queries: Weakly semi-supervised object de- tection by points,” in IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , 2021, pp. 8819–

    2021

  74. [82]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kir- illov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European Conference on Com- puter Vision, 2020. 3

    2020

  75. [83]

    Group r-cnn for weakly semi-supervised object detection with points,

    S. Zhang, Z. Yu, L. Liu, X. Wang, A. Zhou, and K. Chen, “Group r-cnn for weakly semi-supervised object detection with points,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 9407–9416. 3

    2022

  76. [84]

    Object localization under single coarse point supervision,

    X. Yu, P. Chen, D. Wu, N. Hassan, G. Li, J. Yan, H. Shi, Q. Ye, and Z. Han, “Object localization under single coarse point supervision,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 4868–4877. 3

    2022

  77. [85]

    Point2mask: Point-supervised panoptic seg- mentation via optimal transport,

    W. Li, Y . Yuan, S. Wang, J. Zhu, J. Li, J. Liu, and L. Zhang, “Point2mask: Point-supervised panoptic seg- mentation via optimal transport,” in IEEE International Conference on Computer Vision , 2023. 3, 11

    2023

  78. [86]

    Pointobb: Learning oriented object detection via single point supervision,

    J. Luo, X. Yang, Y . Yu, Q. Li, J. Yan, and Y . Li, “Pointobb: Learning oriented object detection via single point supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2024, pp. 16 730–16 740. 3, 11, 12 JOURNAL OF LATEX CLASS FI...

    2024

  79. [87]

    P2rbox: Point prompt oriented object detection with SAM,

    G. Cao, X. Yu, W. Yu, X. Han, X. Yang, G. Li, J. Jiao, and Z. Han, “P2rbox: Point prompt oriented object detection with SAM,” arXiv preprint arXiv:2311.13128,

    arXiv

  80. [88]

    Pointobb- v2: Towards simpler, faster, and stronger single point supervised oriented object detection,

    B. Ren, X. Yang, Y . Yu, J. Luo, and Z. Deng, “Pointobb- v2: Towards simpler, faster, and stronger single point supervised oriented object detection,” in International Conference on Learning Representations , 2025. 3, 11, 12

    2025

  81. [89]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778. 4, 7, 9

    2016

  82. [90]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 936–944. 4, 6, 7, 9

    2017

  83. [91]

    Phase-shifting coder: Predicting accurate orientation in oriented object detection,

    Y . Yu and F. Da, “Phase-shifting coder: Predicting accurate orientation in oriented object detection,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023. 4, 9, 11

    2023

  84. [92]

    You only look one-level feature,

    Q. Chen, Y . Wang, T. Yang, X. Zhang, J. Cheng, and J. Sun, “You only look one-level feature,” in IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2021, pp. 13 034–13 043. 6, 7

    2021

  85. [93]

    Iou loss for 2d/3d object detection,

    D. Zhou, J. Fang, X. Song, C. Guan, J. Yin, Y . Dai, and R. Yang, “Iou loss for 2d/3d object detection,” in International Conference on 3D Vision (3DV), 2019, pp. 85–94. 6

    2019

  86. [94]

    Rotation-robust intersection over union for 3d object detection,

    Y . Zheng, D. Zhang, S. Xie, J. Lu, and J. Zhou, “Rotation-robust intersection over union for 3d object detection,” in European Conference on Computer Vi- sion, 2020, pp. 464–480. 6

    2020

  87. [95]

    Mmrotate: A rotated object detection bench- mark using pytorch,

    Y . Zhou, X. Yang, G. Zhang, J. Wang, Y . Liu, L. Hou, X. Jiang, X. Liu, J. Yan, C. Lyu, W. Zhang, and K. Chen, “Mmrotate: A rotated object detection bench- mark using pytorch,” in 30th ACM International Con- ference on Multimedia , 2022. 8, 9

    2022

  88. [96]

    Fair1m: A benchmark dataset for fine-grained object recognition in high-resolution remote sensing imagery,

    X. Sun, P. Wang, Z. Yan, F. Xu, R. Wang, W. Diao, J. Chen, J. Li, Y . Feng, T. Xu, M. Weinmann, S. Hinz, C. Wang, and K. Fu, “Fair1m: A benchmark dataset for fine-grained object recognition in high-resolution remote sensing imagery,” ISPRS Journal of Photogram- metry and Remot...

    2022

  89. [97]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in International Conference on Learn- ing Representations, 2018. 9

    2018

  90. [98]

    Unitbox: An advanced object detection network,

    J. Yu, Y . Jiang, Z. Wang, Z. Cao, and T. Huang, “Unitbox: An advanced object detection network,” in 24th ACM International Conference on Multimedia , 2016, pp. 516–520. 9

    2016

  91. [99]

    Shape-adaptive se- lection and measurement for oriented object detection,

    L. Hou, K. Lu, J. Xue, and Y . Li, “Shape-adaptive se- lection and measurement for oriented object detection,” in AAAI Conference on Artificial Intelligence , vol. 36, 2022, pp. 923–932. 11

    2022

  92. [100]

    Beyond bounding-box: Convex-hull feature adaptation for oriented and densely packed object detection,

    Z. Guo, C. Liu, X. Zhang, J. Jiao, X. Ji, and Q. Ye, “Beyond bounding-box: Convex-hull feature adaptation for oriented and densely packed object detection,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 8788–8797. 11

    2021

  93. [101]

    Pointobb-v3: Expanding performance boundaries of single point- supervised oriented object detection,

    P. Zhang, J. Luo, X. Yang, Y . Yu, Q. Li, Y . Zhou, X. Jia, X. Lu, J. Chen, X. Li et al. , “Pointobb-v3: Expanding performance boundaries of single point- supervised oriented object detection,” arXiv preprint arXiv:2501.13898, 2025. 11, 12

    2025 arXiv

  94. [102]

    Point2rbox-v2: Rethinking point- supervised oriented object detection with spatial layout among instances,

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

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.