Residual Objectness for Imbalance Reduction

Bin Luo; Dong Liu; Enhong Chen; Joya Chen; Tong Xu; Xuezheng Peng

REVIEW 3 major objections 4 minor 1 cited by

Residual Objectness for Imbalance Reduction

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper proposes Residual Objectness, a fully learned cascade of objectness refinements that addresses foreground-background imbalance without sampling or reweighting, and reports consistent gains on COCO across three detector families.

desk verdict ResObj is a solid empirical component with ~1 AP gains on three detectors, but the paper doesn't isolate the residual mechanism from the extra objectness head, so the causal story is softer than the title suggests. read the letter →

arxiv 1908.09075 v1 pith:RM74O2TZ submitted 2019-08-24 cs.CV

Joya Chen , Dong Liu , Bin Luo , Xuezheng Peng , Tong Xu , Enhong Chen This is my paper · ORCID

classification cs.CV

keywords objectdetectionclassimbalanceforeground-backgroundobjectnessestimationresiduallearningcascadedrefinementCOCO

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that the extreme foreground-background imbalance that makes object detectors hard to train can be addressed by a learning-based architecture alone, without hand-crafted sampling or reweighting. The proposed Residual Objectness (ResObj) mechanism writes the objectness score, the probability that an anchor box covers an object, as an initial estimate plus a cascade of residual refinements, each produced by a separate subnet. On COCO test-dev, the paper reports that replacing Focal Loss with ResObj improves RetinaNet by 3.6% relative AP, enhancing YOLOv3's objectness improves it by 3.9%, and replacing random sampling in Faster R-CNN's proposal network improves it by 3.2%. If true, this would remove a notoriously hyper-parameter-sensitive component of detector training.

What carries the argument

The central object is Residual Objectness: the objectness estimate is written as $o = o_0 + \sum_{t=1}^T r_t$, where $o_0$ comes from an initial objectness subnet and each $r_t$ from a separate residual subnet. The refinement at step $t$ applies only to anchors whose previous objectness score is above a minimum positive score, so easy negatives are progressively filtered out while positives continue to be improved. The residual subnets are isolated from the objectness subnet in backpropagation, and the final class score is computed as $P(\text{class}|\text{obj}) \times P(\text{obj})$. This sum-of-refinements formulation is what turns objectness estimation into a consecutive refinement procedure.

What would settle it

Train a T=1 ResObj detector exactly as in the paper, and a matched control that adds the same objectness head but combines its logits with the initial objectness by concatenation followed by a linear layer, rather than by residual addition, with the same loss and anchor selection. If the control's AP matches or exceeds the residual version, the residual refinement procedure is not the cause of the improvement.

Watch

Extended reading notes

Core claim

The paper's central discovery is that foreground-background imbalance can be transferred from the multi-class classification subnet to a binary objectness task, and then progressively resolved by refining the objectness score several times. A RetinaNet with a single objectness head and plain cross-entropy, no Focal Loss, reaches 34.2 AP versus 34.1 AP for the Focal Loss version, showing that the imbalance is implicitly alleviated by binary objectness prediction. Adding residual refinement, with the objectness written as $o = o_0 + \sum_{t=1}^T r_t$, raises the positive objectness scores over training and yields the reported gains: RetinaNet-ResObj 37.0 AP versus RetinaNet-FL 35.7 with ResNet-50, YOLOv3-ResObj 34.1 versus 33.0 at 608x608, and FasterRCNN-ResObj 38.4 versus 37.2 with ResNet-50. The claim is that this consecutive refinement, not any external sampling schedule, is what fixes the imbalance.

Load-bearing premise

The paper assumes that the observed AP gains come from the residual refinement formulation, rather than from simply adding extra supervised objectness heads; a control with the same number of added heads but without residual connections is missing, so the active ingredient is not isolated.

Editorial extensions

If this is right

Detectors can drop Focal Loss and related hyper-parameters; the paper's RetinaNet-ResObj keeps the same anchor assignment but uses cross-entropy with an objectness head and residual refinement.
The same residual objectness module can be plugged into a one-stage detector (YOLOv3) and a two-stage detector (Faster R-CNN's RPN), improving both by a similar margin.
The cost of the added heads is small: the paper reports inference speed drops by roughly 0.1-0.2 FPS when moving from T=0 to T=2.
Because the final class score multiplies the refined objectness, the mechanism directly sharpens detection confidence, which the paper illustrates with qualitative examples where object scores rise from T=0 to T=2.
The reported gains are consistent across object scales (AP_S, AP_M, AP_L), not confined to large objects.

Reading between the lines

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

The step ablation does not include a control with the same number of extra objectness heads but without residual connections, so whether the residual aggregation is the active ingredient, as opposed to extra supervised heads, remains untested.
The same residual-refinement idea could be applied to any binary foreground/background head, including keypoint or center-point heads in anchor-free detectors, though the paper does not test those.
Because the refinement is applied only to anchors whose previous objectness is above a minimum positive score, the procedure implicitly defines an online curriculum; one could compare its effective loss surface to Focal Loss and GHM analytically.
The reported gains are relative (3-4%), i.e. absolute gains of about 1.0-1.3 AP, and the paper's longer-training ResNet-101 runs use scale jitter; whether the margin persists under identical training schedules is not isolated.

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.

Desk Editor's Note

read the letter

ResObj is a new combination of cascaded objectness heads with residual connections and gradient isolation, and it reports a steady ~1 AP gain on COCO across RetinaNet, YOLOv3, and Faster R-CNN. The single-objectness-head ablation matching Focal Loss is an interesting data point on its own.

What the paper does well: the experiments run on public benchmarks, the gains are consistent across detector families, and the ablations cover gradient flow, head placement, and step count. Inference cost is nearly zero. For practitioners who want to avoid tuning sampling heuristics, the recipe is attractive. The citation pattern looks fine.

The soft spots are real but concentrated. The central claim is that the residual refinement procedure drives the gains. The evidence doesn't isolate it: moving from T=0 to T=1 changes three things at once — an extra objectness head, an extra BCE loss on thresholded anchors, and the residual summation. No non-residual control is presented, so the active ingredient is unidentified. That weakens the novelty and generality story but not the empirical result. The 'first learning-based' claim is an overstatement, since prior objectness modules (RON, DSFD, etc.) are themselves learning-based. ominp in Algorithm 1 is never defined. And the 'no hyper-parameter tuning' claim is softened by T and threshold choices.

The Eq. 4-7 analysis only shows a factor-K reduction in negative loss; it doesn't derive AP. That's acceptable as motivation, but the paper should not lean on it.

Who is this for: anyone working on class imbalance or detector design. A serious referee should require the non-residual control and a proper definition of ominp before acceptance, but the empirical result is solid enough to justify referee time.

Send it to review.

Referee Report

3 major / 4 minor

Summary. The paper proposes Residual Objectness (ResObj), a fully learning-based mechanism for foreground-background imbalance in object detection. The method first transfers the multiclass imbalance to a binary objectness head, then cascades multiple residual objectness subnets that iteratively refine the objectness score, with gradients of the residual subnets isolated from the objectness subnet during backpropagation. The authors replace Focal Loss in RetinaNet, the objectness module in YOLOv3, and random sampling in Faster R-CNN's RPN with the residual objectness mechanism, reporting relative AP improvements of 3.6%, 3.9%, and 3.2% on COCO test-dev, respectively. Ablations examine gradient flow, residual-head placement, and the number of residual steps T.

Significance. The empirical claim is practically valuable: if the residual-refinement mechanism is indeed the active ingredient, ResObj offers a simple, end-to-end trainable alternative to hand-designed sampling/reweighting schemes, with consistent gains across both one-stage and two-stage detectors and negligible inference overhead. The paper ships solid public-benchmark results (COCO test-dev) and a useful ablation set, including a T=0 control showing that the objectness module alone matches Focal Loss and separate ablations for gradient isolation and head placement. However, the central attribution of the gains to the residual formulation is not yet established, and the training procedure depends on an undefined threshold. These issues are fixable but require additional experiments and reporting.

major comments (3)

[Section 4.2, Table 2(c)] The step ablation does not isolate the residual-refinement mechanism. Moving from T=0 (34.1 AP) to T=1 (35.2 AP) simultaneously adds an extra 3x3-convolution objectness head, an extra binary cross-entropy loss on anchors selected by the ominp threshold, and changes the inference score to sigmoid(o0+r1). Since no control with a second independent objectness head (e.g., a head trained with its own binary loss but combined multiplicatively or by summation without the residual framing) is reported, the observed +1.1 AP cannot be attributed to the residual formulation specifically. This is load-bearing because the paper's novelty and generality claims—'consecutive refinement' and 'progressively addressing the imbalance'—depend on the residual mechanism being the active ingredient, not merely on adding an extra supervised objectness head.
[Algorithm 1, line 5; Section 4.2] The threshold ominp_{t-1} is undefined. Algorithm 1 uses it to decide which anchors contribute to the residual loss, and the paper claims the method 'rarely requires hyper-parameters tuning,' but no definition, estimation procedure, or reported values are given. Without this information, the training procedure is not reproducible, and it remains possible that a hand-tuned selection rule carries part of the observed effect. Please specify how ominp is computed (e.g., a percentile of positive objectness scores, a fixed value, or a moving statistic) and report its value or schedule for each detector.
[Section 4.2, Table 2(c); Section 1] The claim that the method avoids hyper-parameter tuning is overstated. The number of residual steps T is an architecture hyper-parameter, the paper's own text says the T=2 result is achieved 'with simple hyper-parameters tuning,' and Table 1 shows that the RetinaNet-Obj baseline's AP varies from 33.1 to 34.1 depending on the inference and NMS thresholds. Please state exactly which hyper-parameters are tuned in the reported results and how sensitive the gains are to the inference threshold, since this directly qualifies the 'no hand-crafted sampling/reweighting' and 'fully learning-based' contributions.

minor comments (4)

[Equation (1)] Equation (1) has an unmatched opening parenthesis after 'FocalLoss'; the summation expression is missing a closing parenthesis. This should be corrected for readability.
[Section 4.3, Table 3] The text states that YOLOv3-ResObj has '1.1 AP (relative 3.9%) lower than us at both 320x320 and 608x608 scales,' but Table 3 shows 28.2 to 29.3 (relative 3.9%) at 320x320 and 33.0 to 34.1 (relative 3.3%) at 608x608. Please correct the inconsistent relative-gain statement.
[Section 4.1 vs Table 3] The ablation tables use a 600-pixel image scale, while Table 3 reports results at 1333x800 for RetinaNet and Faster R-CNN. The text says 'we use a 600 pixel train and test image scale in all ablation studies,' but it would help to explicitly state the scale used for each table, especially because the vanilla RetinaNet baseline differs between Table 2 (34.2 AP) and Table 3 (35.7 AP).
[Section 2] The related-work comparison states that existing cascaded architectures 'are not designed for one-stage detectors,' but C-RPN is discussed as an exception; the wording in the 'Comparison and Difference' paragraph ('most of them... are only applicable for the per-region stage, whereas the only exception C-RPN is designed for object tracking') is slightly confusing and should be clarified.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the AP improvements are measured on COCO, and the loss decomposition in Eqs. (4)-(7) is an analytic observation, not a prediction forced by construction.

full rationale

The paper's central contributions are empirical: Residual Objectness is defined as cascaded objectness modules with residual connections, and its effectiveness is demonstrated by COCO test-dev AP gains against published baselines (RetinaNet, YOLOv3, Faster R-CNN). The only analytical part, Eqs. (4)-(7), compares the negative-loss contribution of Focal Loss versus an objectness loss and shows the objectness formulation reduces the negative loss by roughly a factor of K; this is a loss-level observation and does not by itself entail any AP number. No parameter is fitted to a subset and then reported as a prediction: the ablations vary T and gradient flow, and the final APs are direct benchmark measurements. No load-bearing conclusion is imported from a self-citation; references to Focal Loss, Cascade R-CNN, etc. are standard external context. The main caveat, that the step ablation in Table 2(c) changes capacity, extra loss, and residual aggregation simultaneously, concerns causal attribution of the improvement, not circularity, since the improvement is still measured rather than derived from the method's own assumptions.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The method relies on standard detection training assumptions (anchor assignment, FPN, BCE) and on two underspecified or empirically tuned components: the per-step threshold in Algorithm 1 and the number of residual steps. No new physical entities are introduced.

free parameters (3)

Number of residual steps T = 2
Selected by ablations (T=1,2,3); T=2 chosen as best in Section 4.2, Steps table. Paper claims no laborious tuning, but the number of modules is still an architecture choice.
Inference objectness threshold = 0.05
Table 1 shows RetinaNet-Obj result varies with threshold; 0.05 used for final models per text.
NMS threshold = 0.50
Tuned in Table 1 (0.45 vs 0.50); the paper selects 0.50.

assumptions (3)

domain assumption Anchor assignment with IoU 0.5 for foreground and [0, 0.4) for background is used without reweighting, and all unassigned anchors incur only objectness loss.
Section 3.1; standard RetinaNet setup, but the no-loss-for-class/box on background is part of the method.
domain assumption The minimum positive score ominp in Algorithm 1 can be computed per batch and contains all positives plus hard negatives.
Section 3.2, Algorithm 1 line 5; the paper does not define how this score is computed, yet this threshold gates the residual supervision.
ad hoc to paper Adding objectness subnets on top of the class head and isolating gradients from residual to objectness subnets is beneficial.
Section 4.2, Gradient Flow; the design choices are justified empirically, not derived.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Residual Objectness for Imbalance Reduction." pith.science (2026). https://pith.science/paper/RM74O2TZ

@misc{pith2026190809075,
  author       = {Pith},
  title        = {Pith review of: Residual Objectness for Imbalance Reduction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RM74O2TZ}},
  note         = {Machine review of arXiv:1908.09075}
}

read the original abstract

For a long time, object detectors have suffered from extreme imbalance between foregrounds and backgrounds. While several sampling/reweighting schemes have been explored to alleviate the imbalance, they are usually heuristic and demand laborious hyper-parameters tuning, which is hard to achieve the optimality. In this paper, we first reveal that such the imbalance could be addressed in a learning-based manner. Guided by this illuminating observation, we propose a novel Residual Objectness (ResObj) mechanism that addresses the imbalance by end-to-end optimization, while no further hand-crafted sampling/reweighting is required. Specifically, by applying multiple cascaded objectness-related modules with residual connections, we formulate an elegant consecutive refinement procedure for distinguishing the foregrounds from backgrounds, thereby progressively addressing the imbalance. Extensive experiments present the effectiveness of our method, as well as its compatibility and adaptivity for both region-based and one-stage detectors, namely, the RetinaNet-ResObj, YOLOv3-ResObj and FasterRCNN-ResObj achieve relative 3.6%, 3.9%, 3.2% Average Precision (AP) improvements compared with their vanilla models on COCO, respectively.

Figures

Figures reproduced from arXiv: 1908.09075 by the authors.

**Figure 2.** Average objectness scores on COCO minival vary with training iterations. (a) Varying threshold for RetinaNet-FL Inference NMS AP AP50 AP75 0.100 0.45 34.0 53.0 36.1 0.050 0.50 34.2 53.1 36.5 0.010 0.50 34.2 53.1 36.5 0.005 0.50 34.2 53.1 36.5 0.001 0.50 34.2 53.1 36.5 (b) Varying threshold for RetinaNet-Obj Inference NMS AP AP50 AP75 0.100 0.45 33.1 51.8 35.4 0.050 0.45 33.6 52.2 36.0 0.010 0.45 33.9 52.6 36.3 0.005… view at source ↗

**Figure 3.** Pipeline of RetinaNet-FL and RetinaNet-ResObj. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Average objectness scores on COCO minival vary with training iterations. The results are achieved by RetinaNet-ResObj of T = 2 model. Visualization. As shown in [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

**Figure 5.** Figure 5: Detection results (cases with score > 0.4 are shown) at different steps by RetinaNetResObj with ResNet-50-FPN. It is explicitly shown (red array) that the object scores are gradually improved. accuracy. Motivated by this, a fully learning-based mechanism termed Resid…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Imbalance Problems in Object Detection: A Review
cs.CV 2019-08 accept novelty 6.0 of 10

A taxonomy and critical review organizing eight object-detection imbalance problems under class, scale, spatial, and objective imbalance.

Reference graph

Works this paper leans on

45 extracted references · 37 canonical work pages · cited by 1 Pith paper

[1]

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

work page 2014
[2]

R. B. Girshick, Fast R-CNN, in: IEEE International Conference on Com- puter Vision, 2015, pp. 1440–1448

work page 2015
[3]

S. Ren, K. He, R. B. Girshick, J. Sun, Faster R-CNN: towards real-time object detection with region proposal networks, IEEE Trans. Pattern Anal. Mach. Intell. 39 (6) (2017) 1137–1149

work page 2017
[4]

J. Dai, Y. Li, K. He, J. Sun, R-FCN: object detection via region-based fully convolutional networks, in: Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems, 2016, pp. 379–387

work page 2016
[5]

K. He, G. Gkioxari, P. Doll´ ar, R. B. Girshick, Mask R-CNN, in: IEEE International Conference on Computer Vision, 2017, pp. 2980–2988

work page 2017
[6]

Z. Cai, N. Vasconcelos, Cascade R-CNN: delving into high quality object detection, in: IEEE Conference on Computer Vision and Pattern Recogni- tion, 2018, pp. 6154–6162

work page 2018
[7]

J. Pang, K. Chen, J. Shi, H. Feng, W. Ouyang, D. Lin, Libra R-CNN: towards balanced learning for object detection, CoRR abs/1904.02701. arXiv:1904.02701. URL http://arxiv.org/abs/1904.02701

work page Pith review arXiv 1904
[8]

J. R. R. Uijlings, K. E. A. van de Sande, T. Gevers, A. W. M. Smeulders, Selective search for object recognition, International Journal of Computer Vision 104 (2) (2013) 154–171

work page 2013

Show all 45 references

[9]

C. L. Zitnick, P. Doll´ ar, Edge boxes: Locating object proposals from edges, in: 13th European Conference on Computer Vision, 2014, pp. 391–405. 20

2014
[10]

Redmon, S

J. Redmon, S. K. Divvala, R. B. Girshick, A. Farhadi, You only look once: Uniﬁed, real-time object detection, in: IEEE Conference on Computer Vi- sion and Pattern Recognition, 2016, pp. 779–788

2016
[11]

W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. E. Reed, C. Fu, A. C. Berg, SSD: single shot multibox detector, in: 14th European Conference on Computer Vision, 2016, pp. 21–37

2016
[12]

Redmon, A

J. Redmon, A. Farhadi, YOLO9000: better, faster, stronger, in: IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 6517– 6525

2017
[13]

T. Lin, P. Goyal, R. B. Girshick, K. He, P. Doll´ ar, Focal loss for dense object detection, in: IEEE International Conference on Computer Vision, 2017, pp. 2999–3007

2017
[14]

Zhang, L

S. Zhang, L. Wen, X. Bian, Z. Lei, S. Z. Li, Single-shot reﬁnement neural network for object detection, in: IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 4203–4212

2018
[15]

Redmon, A

J. Redmon, A. Farhadi, Yolov3: An incremental improvement, CoRR abs/1804.02767. arXiv:1804.02767. URL http://arxiv.org/abs/1804.02767

arXiv
[16]

Huang, Y

L. Huang, Y. Yang, Y. Deng, Y. Yu, Densebox: Unifying landmark local- ization with end to end object detection, CoRR abs/1509.04874. arXiv: 1509.04874. URL http://arxiv.org/abs/1509.04874

arXiv
[17]

H. Law, J. Deng, Cornernet: Detecting objects as paired keypoints, in: 15th European Conference on Computer Vision, 2018, pp. 765–781

2018
[18]

X. Zhou, J. Zhuo, P. Kr¨ ahenb¨ uhl, Bottom-up object detection by grouping extreme and center points, CoRR abs/1901.08043. arXiv:1901.08043. URL http://arxiv.org/abs/1901.08043 21

1901 arXiv
[19]

J. Wang, K. Chen, S. Yang, C. C. Loy, D. Lin, Region proposal by guided anchoring, CoRR abs/1901.03278. arXiv:1901.03278. URL http://arxiv.org/abs/1901.03278

1901 arXiv
[20]

T. Kong, F. Sun, H. Liu, Y. Jiang, J. Shi, Foveabox: Beyond anchor-based object detector, CoRR abs/1904.03797. arXiv:1904.03797. URL http://arxiv.org/abs/1904.03797

1904 arXiv
[21]

X. Zhou, D. Wang, P. Kr¨ ahenb¨ uhl, Objects as points, CoRR abs/1904.07850. arXiv:1904.07850. URL http://arxiv.org/abs/1904.07850

1904 arXiv
[22]

B. Li, Y. Liu, X. Wang, Gradient harmonized single-stage detector, in: The Thirty-Third AAAI Conference on Artiﬁcial Intelligence, 2019, pp. 8577–8584

2019
[23]

Shrivastava, A

A. Shrivastava, A. Gupta, R. B. Girshick, Training region-based object detectors with online hard example mining, in: IEEE Conference on Com- puter Vision and Pattern Recognition, 2016, pp. 761–769

2016
[24]

T. Kong, F. Sun, A. Yao, H. Liu, M. Lu, Y. Chen, RON: reverse connection with objectness prior networks for object detection, in: IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 5244–5252

2017
[25]

C. Chi, S. Zhang, J. Xing, Z. Lei, S. Z. Li, X. Zou, Selective reﬁnement network for high performance face detection, in: The Thirty-Third AAAI Conference on Artiﬁcial Intelligence, 2019, pp. 8231–8238

2019
[26]

J. Li, Y. Wang, C. Wang, Y. Tai, J. Qian, J. Yang, C. Wang, J. Li, F. Huang, DSFD: dual shot face detector, CoRR abs/1810.10220. arXiv: 1810.10220. URL http://arxiv.org/abs/1810.10220

arXiv
[27]

P. A. Viola, M. J. Jones, Robust real-time face detection, International Journal of Computer Vision 57 (2) (2004) 137–154. 22

2004
[28]

P. F. Felzenszwalb, R. B. Girshick, D. A. McAllester, Cascade object de- tection with deformable part models, in: IEEE Conference on Computer Vision and Pattern Recognition, 2010, pp. 2241–2248

2010
[29]

Everingham, L

M. Everingham, L. J. V. Gool, C. K. I. Williams, J. M. Winn, A. Zisserman, The pascal visual object classes (VOC) challenge, International Journal of Computer Vision 88 (2) (2010) 303–338

2010
[30]

T. Lin, M. Maire, S. J. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll´ ar, C. L. Zitnick, Microsoft COCO: common objects in context, in: 13th Eu- ropean Conference on Computer Vision, 2014, pp. 740–755

2014
[31]

Alexe, T

B. Alexe, T. Deselaers, V. Ferrari, Measuring the objectness of image win- dows, IEEE Trans. Pattern Anal. Mach. Intell. 34 (11) (2012) 2189–2202

2012
[32]

Alexe, T

B. Alexe, T. Deselaers, V. Ferrari, What is an object?, in: IEEE Conference on Computer Vision and Pattern Recognition, 2010, pp. 73–80

2010
[33]

Cheng, Y

M. Cheng, Y. Liu, W. Lin, Z. Zhang, P. L. Rosin, P. H. S. Torr, BING: bina- rized normed gradients for objectness estimation at 300fps, Computational Visual Media 5 (1) (2019) 3–20

2019
[34]

J. Wang, X. Tao, M. Xu, Y. Duan, J. Lu, Hierarchical objectness network for region proposal generation and object detection, Pattern Recognition 83 (2018) 260–272

2018
[35]

B. Yang, J. Yan, Z. Lei, S. Z. Li, CRAFT objects from images, in: IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 6043– 6051

2016
[36]

Najibi, M

M. Najibi, M. Rastegari, L. S. Davis, G-CNN: an iterative grid based object detector, in: IEEE Conference on Computer Vision and Pattern Recogni- tion, 2016, pp. 2369–2377

2016
[37]

Ghodrati, A

A. Ghodrati, A. Diba, M. Pedersoli, T. Tuytelaars, L. V. Gool, Deeppro- posals: Hunting objects and actions by cascading deep convolutional layers, International Journal of Computer Vision 124 (2) (2017) 115–131. 23

2017
[38]

Cheng, Y

K. Cheng, Y. Chen, W. Fang, Improved object detection with iterative localization reﬁnement in convolutional neural networks, IEEE Trans. Cir- cuits Syst. Video Techn. 28 (9) (2018) 2261–2275

2018
[39]

H. Fan, H. Ling, Siamese cascaded region proposal networks for real-time visual tracking, CoRR abs/1812.06148. arXiv:1812.06148. URL http://arxiv.org/abs/1812.06148

arXiv
[40]

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

2016
[41]

T. Lin, P. Doll´ ar, R. B. Girshick, K. He, B. Hariharan, S. J. Belongie, Feature pyramid networks for object detection, in: IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 936–944

2017
[42]

Girshick, I

R. Girshick, I. Radosavovic, G. Gkioxari, P. Doll´ ar, K. He, Detectron, https://github.com/facebookresearch/detectron (2018)

2018
[43]

Redmon, Darknet: Open source neural networks in c, http://pjreddie

J. Redmon, Darknet: Open source neural networks in c, http://pjreddie. com/darknet/ (2013–2016)

2013
[44]

Russakovsky, J

O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. S. Bernstein, A. C. Berg, F. Li, Imagenet large scale visual recognition challenge, International Journal of Computer Vision 115 (3) (2015) 211–252

2015
[45]

Francisco, G

M. Francisco, G. Ross, maskrcnn-benchmark: Fast, modular ref- erence implementation of Instance Segmentation and Object Detec- tion algorithms in PyTorch, https://github.com/facebookresearch/ maskrcnn-benchmark, accessed: [July 21, 2019] (2018). 24

2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

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

work page 2014

[2] [2]

R. B. Girshick, Fast R-CNN, in: IEEE International Conference on Com- puter Vision, 2015, pp. 1440–1448

work page 2015

[3] [3]

S. Ren, K. He, R. B. Girshick, J. Sun, Faster R-CNN: towards real-time object detection with region proposal networks, IEEE Trans. Pattern Anal. Mach. Intell. 39 (6) (2017) 1137–1149

work page 2017

[4] [4]

J. Dai, Y. Li, K. He, J. Sun, R-FCN: object detection via region-based fully convolutional networks, in: Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems, 2016, pp. 379–387

work page 2016

[5] [5]

K. He, G. Gkioxari, P. Doll´ ar, R. B. Girshick, Mask R-CNN, in: IEEE International Conference on Computer Vision, 2017, pp. 2980–2988

work page 2017

[6] [6]

Z. Cai, N. Vasconcelos, Cascade R-CNN: delving into high quality object detection, in: IEEE Conference on Computer Vision and Pattern Recogni- tion, 2018, pp. 6154–6162

work page 2018

[7] [7]

J. Pang, K. Chen, J. Shi, H. Feng, W. Ouyang, D. Lin, Libra R-CNN: towards balanced learning for object detection, CoRR abs/1904.02701. arXiv:1904.02701. URL http://arxiv.org/abs/1904.02701

work page Pith review arXiv 1904

[8] [8]

J. R. R. Uijlings, K. E. A. van de Sande, T. Gevers, A. W. M. Smeulders, Selective search for object recognition, International Journal of Computer Vision 104 (2) (2013) 154–171

work page 2013