Pith. sign in

REVIEW 4 major objections 5 minor 48 references

C-RPNs: Promoting Object Detection in real world via a Cascade Structure of Region Proposal Networks

T0 review · 4 major / 5 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 A four-stage cascade of region proposal networks rejects easy samples stage by stage, claiming wild-scene bird detection AP rises from 59.3% to 70.3% on BSBDV 2017.

desk verdict A cascade of RPNs with easy-sample rejection is a sensible, incremental idea, but the headline 11-point wild-bird gain is weakened by test-set hyperparameter tuning and a contradicting loss-weighting formula. read the letter →

arxiv 1908.06665 v1 pith:YOCAKE4U submitted 2019-08-19 cs.CV

Dongming Yang , YueXian Zou , Jian Zhang , Ge Li This is my paper · ORCID
classification cs.CV
keywords objectdetectionhardexampleminingcascadenetworkregionproposaldataimbalancereal-worldfeaturechainscore
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

Object detectors trained on clean benchmark photos stumble in the wild, where small, partially occluded objects sit inside huge easy backgrounds. This paper tries to fix that by moving hard-sample mining into the region proposal stage: instead of one proposal network, it chains four, each rejecting confident easy samples so later stages specialize on difficult ones. The paper claims that this cascade, aided by a feature chain, a score chain, and a staged loss, raises average precision on wild-scene bird detection from 59.3% (Faster R-CNN) to 70.3% on BSBDV 2017 and improves pedestrian detection on Caltech and CityPersons, while staying competitive on the standard Pascal VOC benchmark. If true, it shows that data-imbalance losses in real-world detection can be recovered at proposal time without new annotations.

What carries the argument

The carrying mechanism is the four-stage cascade built on VGG16 features (Conv4_2, Conv4_3, Conv5_2, Conv5_3, with 2x2 average pooling to align resolutions). At each stage t, the feature chain computes $h_t = \lambda_f h_{t-1} \oplus \lambda_p f_t$ for $t>1$ and the score chain computes $s_t = \lambda_f s_{t-1} + \lambda_p c_t$; the cascade loss is $L_{cls} = -\sum_{t=1}^T \alpha_t \mu_t \log(s_{t,k^*})$ with $\alpha_t$ increasing tenfold per stage and $\mu_t = \prod_{i<t} [s_{i,k^*} < r]$ enforcing rejection of easy samples. This object is what transfers hard-sample mining into proposal generation and lets all classifiers be learned jointly by backpropagation.

What would settle it

Run C-RPNs on a new real-world detection set with many small, occluded instances (for example, drone or surveillance imagery) without retuning r and the fusion weights, and count how many ground-truth positives are rejected at stages 1-3. If AP gains over a single-stage RPN disappear, or if the rejected positives include hard true objects, the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that a cascade of four region proposal networks, each with its own binary classifier and a shared reject threshold, can mine hard samples at the proposal stage rather than at the final classification stage. C-RPNs discards easy samples early—anything scored above r=0.99 by an earlier stage stops advancing—so stages 2-4 focus on the remaining hard examples, and the final stage also performs bounding-box regression. With a feature chain that fuses previous-stage and current-stage convolutional features, a score chain that accumulates classifier outputs, and a cascade loss whose per-stage weights increase tenfold, the model is trained end-to-end on top of Faster R-CNN. On BSBDV 2017, a wild-scene bird dataset, C-RPNs reports 70.3% AP versus 59.3% for Faster R-CNN, and it reports 76.4% mAP on Pascal VOC 2007 test, with gains on Caltech and CityPersons. That is the paper's central discovery and claim.

Load-bearing premise

Everything rests on two tuning choices—the reject threshold r=0.99 and fusion weights $\lambda_f=0.1$, $\lambda_p=0.9$—selected by grid search on the same BSBDV 2017 test images used for the headline AP; if those choices do not generalize to other real-world scenes, or if early rejection throws away genuinely hard objects, the claimed improvement does not transfer.

Editorial extensions

If this is right

  • Mining hard samples during proposal generation lets a two-stage detector train on harder, cleaner proposal distributions without any annotation cost beyond the original boxes.
  • The cascade's stage-wise batch sizes shrink from 1024 to 256, so computation focuses on the samples that survive rejection; later classifiers see fewer, harder proposals.
  • The staged loss trains all four classifiers jointly through backpropagation, with deeper stages weighted tenfold more, allowing the whole C-RPN detector to be optimized end-to-end.
  • On Pascal VOC 2007 test, C-RPNs reaches 76.4% mAP (3.2 points above Faster R-CNN), showing the real-world-oriented design also helps standard benchmark detection.
  • Error analysis with the tool from [47] shows fewer background and localization false positives than Faster R-CNN on animals and vehicles, and higher normalized AP across occlusion, truncation, box size, aspect ratio, viewpoint, and part visibility.

Reading between the lines

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

  • Extension the authors do not draw: the reject threshold r=0.99 and fusion weight $\lambda_f=0.1$ are fixed settings; a natural test is a schedule that lowers r as stages deepen or lets a small random fraction of high-confidence proposals bypass rejection, so very hard positives are not lost.
  • C-RPNs' transfer gain over FPN is larger on wild birds (3.4 AP) than its margin on VOC, so the same cascade could be evaluated as a plug-in module for one-stage detectors or feature-pyramid detectors to see whether proposal-stage mining, rather than the extra stages alone, is the active ingredient.
  • Since the final detector still confuses similar object categories, the cascade's binary object/background mining leaves room for a class-aware hard-sample strategy at the RoI head; combining C-RPNs with category-level hard mining is a testable next step.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 proposes C-RPNs, a four-stage cascade of region proposal networks built on Faster R-CNN with VGG16, which mines hard samples during proposal generation by rejecting easy samples at early stages using a threshold r. A feature chain and a score chain propagate previous-stage information into later stages, and a staged loss with weights alpha_t and indicator variables mu_t trains all cascade classifiers jointly. The method is evaluated on PASCAL VOC 2007 test, BSBDV 2017, Caltech Pedestrian Benchmark, and CityPersons, reporting 76.4% mAP on VOC 2007 and 70.3% AP on BSBDV 2017, outperforming the Faster R-CNN baseline by 11 points on the bird dataset. Ablations on BSBDV 2017 examine the number of stages and the contribution of the feature and score chains.

Significance. If the reported improvements are reproducible with a properly controlled evaluation, C-RPNs would be a useful empirical contribution to object detection under data imbalance, with the cascade-on-RPN idea, feature/score chains, and staged-loss training being clearly described and ablated. The paper also provides multiple datasets, qualitative examples, and a detailed error analysis using the Hoiem et al. diagnostic tools, which are strengths. However, the headline real-world result is currently selected by grid search on the BSBDV 2017 test set, and several comparisons are not backbone-controlled, so the significance is conditional on resolving those evaluation issues.

major comments (4)
  1. [Section 4.5 and Table 2] The reject threshold r and fusion weight lambda_f are chosen by one-dimensional grid search on the BSBDV 2017 test set, and the best configuration (70.31% AP) is then reported as the headline result in Table 2; the baselines are not given the same test-set adaptation, so the claimed 11% improvement over Faster R-CNN is not an unbiased comparison. Please evaluate on a held-out validation split or fix the hyperparameters before touching the test set, and report the test result for that fixed configuration.
  2. [Section 3.3, Eq. (5)] Eq. (5) defines alpha_t = alpha_T * 10^(T-t); with T=4 and alpha_T=1, this gives alpha = [1000, 100, 10, 1] for t=1,...,4, meaning shallow stages receive larger weights, which is the opposite of the prose claim that deeper classifiers receive a tenfold increase in weight. The correct formula or the retrained results must be provided, since the exact trained objective is currently ambiguous.
  3. [Tables 1-2] The comparisons in Tables 1 and 2 are not controlled for backbone network: in Table 2, C-RPNs uses VGG16 while RON uses ResNet-101 and R-FCN and FPN use ResNet-50, so the claim that C-RPNs is more competitive in real-world detection is not established on an equal footing. Same-backbone comparisons (e.g., FPN with VGG16 or C-RPNs with ResNet) are needed to support the stated conclusions.
  4. [Section 3.3, Eqs. (4)-(6)] The training loss uses mu_t with the ground-truth class score s(t,k*), but the inference-time rejection criterion is not specified; the paper should state whether rejection at test time uses the object score s(t,1), the maximum score, or some other function, because the cascade's behavior and the meaning of the reported detection results depend on this choice.
minor comments (5)
  1. [Section 3.1] The text contains a typo, 'Con4 3', which should be 'Conv4 3'.
  2. [Section 4.3] The phrase 'verify the efficiency of our approach' in the discussion of pedestrian datasets should likely be 'verify the efficacy' to match the paper's claims.
  3. [Table 3] Table 3 compares only against the Faster R-CNN baseline on Caltech and CityPersons; adding state-of-the-art pedestrian detectors would strengthen the claim of competitive performance on these datasets.
  4. [Section 4.5] The grid-search description in Section 4.5 does not give the ranges or step sizes used for r and lambda_f, which limits reproducibility; please include these details.
  5. [Section 3.2, Eq. (1)] The notation around Eq. (1) is inconsistent: the text says lambda = {lambda_f, lambda_p} and that lambda_f and lambda_p add to 1, but later only lambda_f is mentioned in the grid search; please clarify the parameter notation and the constraint.

Circularity Check

1 steps flagged · score 6.0 of 10

BSBDV 2017 headline AP is the maximum of a test-set grid search, making one central real-world result partially circular; the method's other evaluation benchmarks remain independent.

  1. fitted input called prediction [Section 4.5 (Selection of reject threshold and fusion rate), reported in Table 2 and Abstract]
    "To find the best hyper parameters, empirical tests were conducted using different reject threshold r and fusion rate λf on BSBDV 2017 through one-dimensional grid search. ... reject threshold r=0.99 achieved the best AP of 70.31% when the fusion rate was fixed at 0.1. We then fixed the reject threshold as 0.99 and applied a grid search by changing the fusion rate λf. From Figure 10, the best λf is observed as 0.1 with the AP of 70.31%."

    The 70.3% AP reported for C-RPNs on BSBDV 2017 is the same number selected as the best point of a grid search over r and λf performed on BSBDV 2017; no held-out validation split is described. Thus the headline result is by construction the maximum over the searched hyperparameters rather than an unbiased fixed-configuration test. Comparing this selected maximum against untuned baselines (e.g., Faster R-CNN at 59.3%) is not on equal footing, so the central 'real-world improvement' evidence from BSBDV is partially forced by the tuning procedure.

full rationale

The architecture and loss of C-RPNs are not self-deriving: the feature chain, score chain, cascade loss, and hard-sample rejection are defined from the network's own intermediate outputs, but no equation reduces a predicted quantity to an input by construction. No load-bearing self-citation or imported uniqueness theorem appears; the BSBDV dataset citation [3] is a public benchmark and is not used to justify the method. The main circular element is the test-set grid search: r=0.99 and λf=0.1 are selected on BSBDV 2017, and the same dataset's test AP (70.3%) is then touted as the method's real-world result. That specific number is statistically forced. However, the paper also reports Pascal VOC 2007 (76.4 mAP), Caltech (48.1 AP), CityPersons (51.4 AP), and component ablations that were not used in that grid search, providing independent empirical grounding for the general claim. I also note an internal inconsistency in Eq. (5): with T=4 and α_T=1, α_t=10^{4-t} gives [1000,100,10,1], the opposite of the prose claim that deeper classifiers get tenfold more weight; this is a correctness/ambiguity issue, not circularity. Overall, one headline number reduces by construction, so partial circularity is present but the derivation itself is not globally circular.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central method rests on several free hyperparameters (r, lambda_f/lambda_p, alpha_t, batch sizes, T). The most concerning are r and lambda_f, which are chosen by grid search on the same BSBDV 2017 test set whose AP is then reported as the headline result. The alpha_t formula also contradicts the paper's stated design intent. No new physical or conceptual entities are invented.

free parameters (5)
  • Reject threshold r = 0.99
    Grid-searched on BSBDV 2017 test images to maximize AP (Fig. 10); controls which easy samples are discarded at each stage.
  • Feature fusion weight lambda_f = 0.1
    Grid-searched on BSBDV 2017 test set; lambda_p is set to 0.9 since lambda_f + lambda_p = 1.
  • Cascade loss weights alpha_t = alpha_t = 1 * 10^(4-t) for T=4 (alpha_1=1000, alpha_4=1)
    Chosen 'based on our experience', contradicting the prose that deeper stages get more weight.
  • Per-stage batch sizes = 1024, 768, 512, 256
    Chosen by hand so stage 4 matches Faster R-CNN RPN batch size of 256; no sensitivity analysis.
  • Number of cascade stages T = 4
    Set to 4 because experiments showed 5 stages add little but cost time; no detailed results shown.
assumptions (4)
  • domain assumption Easy samples can be safely discarded without losing hard positive information.
    Stated in Section 1; formalized by threshold r in Eq. (6).
  • domain assumption ImageNet-pretrained VGG16 features transfer to bird and pedestrian detection.
    Used as backbone; standard in the field but not tested for these datasets.
  • domain assumption Joint backpropagation through the cascade loss trains all stages without gradient pathology.
    Assumed; no gradient analysis or learning-curve diagnostics are provided.
  • ad hoc to paper Weighted linear combination of features and scores across stages preserves discriminative information.
    Feature chain Eq. (1) and score chain Eq. (2) use weights fixed by grid search rather than learned or derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of C-RPNs: Promoting Object Detection in real world via a Cascade Structure of Region Proposal Networks." pith.science (2026). https://pith.science/paper/YOCAKE4U

@misc{pith2026190806665,
  author       = {Pith},
  title        = {Pith review of: C-RPNs: Promoting Object Detection in real world via a Cascade Structure of Region Proposal Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YOCAKE4U}},
  note         = {Machine review of arXiv:1908.06665}
}
read the original abstract

Recently, significant progresses have been made in object detection on common benchmarks (i.e., Pascal VOC). However, object detection in real world is still challenging due to the serious data imbalance. Images in real world are dominated by easy samples like the wide range of background and some easily recognizable objects, for example. Although two-stage detectors like Faster R-CNN achieved big successes in object detection due to the strategy of extracting region proposals by region proposal network, they show their poor adaption in real-world object detection as a result of without considering mining hard samples during extracting region proposals. To address this issue, we propose a Cascade framework of Region Proposal Networks, referred to as C-RPNs. The essence of C-RPNs is adopting multiple stages to mine hard samples while extracting region proposals and learn stronger classifiers. Meanwhile, a feature chain and a score chain are proposed to help learning more discriminative representations for proposals. Moreover, a loss function of cascade stages is designed to train cascade classifiers through backpropagation. Our proposed method has been evaluated on Pascal VOC and several challenging datasets like BSBDV 2017, CityPersons, etc. Our method achieves competitive results compared with the current state-of-the-arts and all-sided improvements in error analysis, validating its efficacy for detection in real world.

Figures

Figures reproduced from arXiv: 1908.06665 by the authors.

Figure 1
Figure 1. Examples: (1) 12 images from Pascal VOC (left upper); (2) one littoral bird image [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. An overall of our proposed C-RPNs model. We adopt VGG16 as backbone network. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. We define the number of stages as T and t is the stage index. At stage t, we 9 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (8 more)
Figure 3
Figure 3. Figure 3: The proposed feature chain and score chain of C-RPNs. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 4
Figure 4. Figure 4: Cascade Losses of our proposed C-RPNs. Faster R-CNN [4] is displayed as baseline [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Detection results of Faster R-CNN (row 1) and our proposed C-RPNs (row 2) on [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Detection results of Faster R-CNN (row 1 and row 3) and our proposed C-RPNs [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Analysis of Top-Ranked False Positives. Pie charts: fraction of detections that are [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Characteristics analysis of different bird characteristics on VOC2007 test set: Each [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Summary of Sensitivity and Impact of Object Characteristics. The APN are shown [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Grid search for the best reject threshold and fusion rate. Left: accuracy vs reject [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 41 canonical work pages

  1. [1]

    Everingham, L

    M. Everingham, L. Van Gool, C. K. Williams, J. Winn, A. Zisserman, The pascal visual object classes (voc) challenge, International Journal of Computer Vision 88 (2) (2010) 303–338. 22

    work page 2010

  2. [2]

    T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll´ ar, C. L. Zitnick, Microsoft coco: Common objects in context, in: European Conference on Computer Vision, Springer, 2014, pp. 740–755

    work page 2014

  3. [3]

    W. Guan, Y. Zou, X. Zhou, Multi-scale object detection with feature fusion and region objectness network, in: 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2018, pp. 2596– 2600

    work page 2018

  4. [4]

    S. Ren, K. He, R. B. Girshick, J. Sun, Faster r-cnn: Towards real-time ob- ject detection with region proposal networks, IEEE Transactions on Pattern Analysis and Machine Intelligence 39 (6) (2017) 1137–1149

    work page 2017

  5. [5]

    Dollar, C

    P. Dollar, C. Wojek, B. Schiele, P. Perona, Pedestrian detection: A bench- mark, in: Computer Vision and Pattern Recognition, 2009. IEEE Confer- ence on, 2009, pp. 304–311

    work page 2009

  6. [6]

    Zhang, R

    S. Zhang, R. Benenson, B. Schiele, Citypersons: A diverse dataset for pedestrian detection, in: The IEEE Conference on Computer Vision and Pattern Recognition, Vol. 1, 2017, p. 3

    work page 2017

  7. [7]

    Girshick, J

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

    work page 2014

  8. [8]

    Girshick, Fast r-cnn, in: Proceedings of the IEEE international Confer- ence on Computer Vision, 2015, pp

    R. Girshick, Fast r-cnn, in: Proceedings of the IEEE international Confer- ence on Computer Vision, 2015, pp. 1440–1448

    work page 2015

Show all 48 references
  1. [9]

    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, 2016, pp. 379–387

    2016

  2. [10]

    Gidaris, N

    S. Gidaris, N. Komodakis, Object detection via a multi-region and semantic segmentation-aware cnn model, in: Proceedings of the IEEE International Conference on Computer Vision, 2015, pp. 1134–1142. 23

    2015

  3. [11]

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, A. C. Berg, Ssd: Single shot multibox detector, in: European Conference on Computer Vision, Springer, 2016, pp. 21–37

    2016

  4. [12]

    Redmon, S

    J. Redmon, S. Divvala, R. Girshick, A. Farhadi, You only look once: Uni- fied, real-time object detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 779–788

    2016

  5. [13]

    T.-Y. Lin, P. Doll´ ar, R. Girshick, K. He, B. Hariharan, S. Belongie, Fea- ture pyramid networks for object detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Vol. 1, 2017, p. 4

    2017

  6. [14]

    Zhang, L

    L. Zhang, L. Lin, X. Liang, K. He, Is faster r-cnn doing well for pedestrian detection?, in: European Conference on Computer Vision, Springer, 2016, pp. 443–457

    2016

  7. [15]

    J. Dai, H. Qi, Y. Xiong, Y. Li, G. Zhang, H. Hu, Y. Wei, Deformable con- volutional networks, in: Proceedings of the IEEE International Conference on Computer Vision, 2017, pp. 764–773

    2017

  8. [16]

    Simonyan, A

    K. Simonyan, A. Zisserman, Very deep convolutional networks for large- scale image recognition, in: Computer Science, 2014

    2014

  9. [17]

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

    2016

  10. [18]

    Szegedy, W

    C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, A. Rabinovich, Going deeper with convolutions, in: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition, 2015, pp. 1–9

    2015

  11. [19]

    S. Bell, C. Lawrence Zitnick, K. Bala, R. Girshick, Inside-outside net: De- tecting objects in context with skip pooling and recurrent neural networks, 24 in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 2874–2883

    2016

  12. [20]

    Zagoruyko, N

    S. Zagoruyko, N. Komodakis, Wide residual networks, in: British Machine Vision Conference, 2016

    2016

  13. [21]

    S. Liu, D. Huang, et al., Receptive field block net for accurate and fast object detection, in: Proceedings of the European Conference on Computer Vision (ECCV), 2018, pp. 385–400

    2018

  14. [22]

    C. Peng, T. Xiao, Z. Li, Y. Jiang, X. Zhang, K. Jia, G. Yu, J. Sun, Megdet: A large mini-batch object detector, Computer Vision and Pattern Recog- nition

  15. [23]

    Y. Liu, R. Wang, S. Shan, X. Chen, Structure inference net: Object detec- tion using scene-level context and instance-level relationships, in: Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 6985–6994

    2018

  16. [24]

    Sung, Learning and example selection for object and pattern detec- tion, 1996

    K.-K. Sung, Learning and example selection for object and pattern detec- tion, 1996

    1996

  17. [25]

    K. He, X. Zhang, S. Ren, J. Sun, Spatial pyramid pooling in deep con- volutional networks for visual recognition, in: European Conference on Computer Vision, Springer, 2014, pp. 346–361

    2014

  18. [26]

    Simoserra, E

    E. Simoserra, E. Trulls, L. Ferraz, I. Kokkinos, F. Morenonoguer, Fracking deep convolutional image descriptors, in: Computer Science, 2015

    2015

  19. [27]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Online batch selection for faster training of neural networks, arXiv preprint arXiv:1511.06343

    arXiv

  20. [28]

    Shrivastava, A

    A. Shrivastava, A. Gupta, R. Girshick, Training region-based object de- tectors with online hard example mining, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 761– 769. 25

    2016

  21. [29]

    T.-Y. Lin, P. Goyal, R. Girshick, K. He, P. Doll´ ar, Focal loss for dense object detection, IEEE transactions on pattern analysis and machine intel- ligence

  22. [30]

    P. F. Felzenszwalb, R. B. Girshick, D. McAllester, Cascade object detection with deformable part models, in: Computer vision and Pattern Recogni- tion, 2010 IEEE Conference on, IEEE, 2010, pp. 2241–2248

    2010

  23. [31]

    Doll´ ar, R

    P. Doll´ ar, R. Appel, S. Belongie, P. Perona, Fast feature pyramids for object detection, IEEE Transactions on Pattern Analysis and Machine Intelligence 36 (8) (2014) 1532–1545

    2014

  24. [32]

    R. Xiao, L. Zhu, H.-J. Zhang, Boosting chain learning for object detec- tion, in: Computer Vision, 2003. Proceedings. Ninth IEEE International Conference on, IEEE, 2003, pp. 709–715

    2003

  25. [33]

    Bourdev, J

    L. Bourdev, J. Brandt, Robust object detection via soft cascade, in: Com- puter Vision and Pattern Recognition, 2005. IEEE Computer Society Con- ference on, Vol. 2, IEEE, 2005, pp. 236–243

    2005

  26. [34]

    Ouyang, X

    W. Ouyang, X. Wang, X. Zeng, S. Qiu, P. Luo, Y. Tian, H. Li, S. Yang, Z. Wang, C.-C. Loy, et al., Deepid-net: Deformable deep convolutional neu- ral networks for object detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 2403–2412

    2015

  27. [35]

    H. Li, Z. Lin, X. Shen, J. Brandt, G. Hua, A convolutional neural network cascade for face detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 5325–5334

    2015

  28. [36]

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

    2016

  29. [37]

    H. Qin, J. Yan, X. Li, X. Hu, Joint training of cascaded cnn for face detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 3456–3465. 26

    2016

  30. [38]

    Z. Cai, N. Vasconcelos, Cascade r-cnn: Delving into high quality object detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 6154–6162

    2018

  31. [39]

    C.-Y. Fu, W. Liu, A. Ranga, A. Tyagi, A. C. Berg, Dssd: Deconvolutional single shot detector, arXiv preprint arXiv:1701.06659

    arXiv

  32. [40]

    D. Yang, J. Zhang, S. Xu, S. Ge, G. H. Kumar, X. Zhang, Real-time pedes- trian detection via hierarchical convolutional feature, Multimedia Tools and Applications 77 (19) (2018) 25841–25860

    2018

  33. [41]

    A. Ess, B. Leibe, K. Schindler, L. Van Gool, A mobile vision system for ro- bust multi-person tracking, in: Computer Vision and Pattern Recognition,

  34. [42]

    Wojek, S

    C. Wojek, S. Walk, B. Schiele, Multi-cue onboard pedestrian detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2009, pp. 794–801

    2009

  35. [43]

    Cordts, M

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, B. Schiele, The cityscapes dataset for semantic urban scene understanding, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 3213–3223

    2016

  36. [44]

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

    2009

  37. [45]

    Abadi, A

    M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, M. Devin, et al., Tensorflow: Large-scale ma- chine learning on heterogeneous distributed systems, arXiv: Distributed, Parallel, and Cluster Computing

  38. [46]

    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, Vol. 1, 2017, p. 2. 27

    2017

  39. [47]

    Hoiem, Y

    D. Hoiem, Y. Chodpathumwan, Q. Dai, Diagnosing error in object detec- tors, in: European Conference on Computer Vision, Springer, 2012, pp. 340–353. 28

    2012

  40. [2008]

    IEEE Conference on, IEEE, 2008, pp. 1–8

    2008

Pith tools

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