Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
110 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Dynamic Loss Decay based Robust Oriented Object Detection on Remote Sensing Images with Noisy Labels (2405.09024v1)

Published 15 May 2024 in cs.CV

Abstract: The ambiguous appearance, tiny scale, and fine-grained classes of objects in remote sensing imagery inevitably lead to the noisy annotations in category labels of detection dataset. However, the effects and treatments of the label noises are underexplored in modern oriented remote sensing object detectors. To address this issue, we propose a robust oriented remote sensing object detection method through dynamic loss decay (DLD) mechanism, inspired by the two phase early-learning'' andmemorization'' learning dynamics of deep neural networks on clean and noisy samples. To be specific, we first observe the end point of early learning phase termed as EL, after which the models begin to memorize the false labels that significantly degrade the detection accuracy. Secondly, under the guidance of the training indicator, the losses of each sample are ranked in descending order, and we adaptively decay the losses of the top K largest ones (bad samples) in the following epochs. Because these large losses are of high confidence to be calculated with wrong labels. Experimental results show that the method achieves excellent noise resistance performance tested on multiple public datasets such as HRSC2016 and DOTA-v1.0/v2.0 with synthetic category label noise. Our solution also has won the 2st place in the "fine-grained object detection based on sub-meter remote sensing imagery" track with noisy labels of 2023 National Big Data and Computing Intelligence Challenge.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (50)
  1. Sample-wise label confidence incorporation for learning with noisy labels. In ICCV, pages 1823–1832, 2023.
  2. A closer look at memorization in deep networks. In ICML, pages 233–242. PMLR, 2017.
  3. Correcting imprecise object locations for training object detectors in remote sensing applications. Remote Sensing, 13(24):4962, 2021.
  4. Weakly supervised deep detection networks. In CVPR, pages 2846–2854, 2016.
  5. Ao2-detr: Arbitrary-oriented object detection transformer. IEEE TCSVT, 2022.
  6. 2023 National Big Data and Computing Intelligence Challenge. https://www.datafountain.cn/competitions/637, 2023. [Online; accessed 17-November-2023].
  7. Learning roi transformer for oriented object detection in aerial images. In CVPR, pages 2849–2858, 2019.
  8. Object detection in aerial images: A large-scale benchmark and challenges. IEEE TPAMI, pages 1–1, 2021.
  9. Generalized jensen-shannon divergence loss for learning with noisy labels. NeurIPS, 34:30284–30297, 2021.
  10. Ot-filter: An optimal transport filter for learning with noisy labels. In CVPR, pages 16164–16174, 2023.
  11. Beyond bounding-box: Convex-hull feature adaptation for oriented and densely packed object detection. In CVPR, pages 8792–8801, 2021.
  12. Co-teaching: Robust training of deep neural networks with extremely noisy labels. NeurIPS, 31, 2018.
  13. Align deep features for oriented object detection. IEEE Trans. Geosci. Remo. Sens., 60:1–11, 2021a.
  14. Redet: A rotation-equivariant detector for aerial object detection. In CVPR, pages 5–7, 14, 2021b.
  15. Probability differential-based class label noise purification for object detection in aerial images. IEEE GRSL, 19:1–5, 2022.
  16. A general gaussian heatmap label assignment for arbitrary-oriented object detection. IEEE TIP, 31:1895–1910, 2022.
  17. Disentangling human error from the ground truth in segmentation of medical images. In NeurIPS. ACL, 2021.
  18. Multilayer spectral–spatial graphs for label noisy robust hyperspectral image classification. IEEE TNNLS, 33(2):839–852, 2020.
  19. Mentornet: Learning data-driven curriculum for very deep neural networks on corrupted labels. In ICML, pages 2304–2313. PMLR, 2018.
  20. Regularizing neural networks by penalizing confident output distributions. In ICML, 2012.
  21. Towards noise-resistant object detection with noisy annotations. arXiv preprint arXiv:2003.01285, 2020.
  22. Semi-supervised semantic segmentation under label noise via diverse learning groups. In ICCV, pages 1229–1238, 2023a.
  23. Oriented reppoints for aerial object detection. In CVPR, pages 1829–1838, 2022.
  24. Large selective kernel network for remote sensing object detection. In ICCV, pages 16794–16805, 2023b.
  25. Robust object detection with inaccurate bounding boxes. In ECCV, pages 53–69. Springer, 2022a.
  26. Early-learning regularization prevents memorization of noisy labels. NeurIPS, 33:20331–20342, 2020.
  27. Adaptive early-learning correction for segmentation from noisy annotations. In CVPR, pages 2606–2616, 2022b.
  28. A high resolution optical satellite image dataset for ship recognition and some new baselines. In International Conference on Pattern Recognition Applications and Methods, 2017.
  29. Swin transformer: Hierarchical vision transformer using shifted windows. In ICCV, pages 10012–10022, 2021.
  30. Eigen-cam: Class activation map using principal components. In 2020 International Joint Conference on Neural Networks (IJCNN), pages 1–7, 2020.
  31. Hyperspectral classification with noisy label detection via superpixel-to-pixel weighting distance. IEEE Trans. Geosci. Remo. Sens., 58(6):4116–4131, 2020.
  32. Advancing plain vision transformer toward remote sensing foundation model. IEEE Trans. Geosci. Remo. Sens., 61:1–15, 2023a.
  33. A noise-robust framework for automatic segmentation of covid-19 pneumonia lesions from ct images. IEEE TMI, 39(8):2653–2663, 2020.
  34. Oriented object detection in optical remote sensing images: A survey. arXiv preprint arXiv:2302.10473, 2023b.
  35. Spatial self-distillation for object detection with inaccurate bounding boxes. In ICCV, pages 6855–6865, 2023.
  36. Dota: A large-scale dataset for object detection in aerial images. In CVPR, 2018.
  37. Oriented r-cnn for object detection. In ICCV, pages 3520–3529, 2021.
  38. Dynamic coarse-to-fine learning for oriented tiny object detection. In CVPR, pages 7318–7328, 2023.
  39. Training robust object detectors from noisy category labels and imprecise bounding boxes. IEEE TIP, 30:5782–5792, 2021.
  40. Dense label encoding for boundary discontinuity free rotation detection. In CVPR, pages 15819–15829, 2021a.
  41. R3det: Refined single-stage detector with feature refinement for rotating object. In AAAI, pages 3163–3171, 2021b.
  42. Rethinking rotated object detection with gaussian wasserstein distance loss. In ICML, pages 11830–11841. PMLR, 2021c.
  43. Learning high-precision bounding box for rotated object detection via kullback-leibler divergence. NeurIPS, 34:18381–18394, 2021d.
  44. Phase-shifting coder: Predicting accurate orientation in oriented object detection. In CVPR, pages 13354–13363, 2023.
  45. Ars-detr: Aspect ratio sensitive oriented object detection with transformer. arXiv preprint arXiv:2303.04989, 2023.
  46. Understanding deep learning (still) requires rethinking generalization. Communications of the ACM, 64(3):107–115, 2021.
  47. Learning to localize objects with noisy labeled instances. In AAAI, pages 9219–9226, 2019.
  48. Triple contrastive representation learning for hyperspectral image classification with noisy labels. IEEE Trans. Geosci. Remo. Sens., 2023.
  49. MMRotate. In ACM MM. ACM, 2022.
  50. Zhi-Hua Zhou. A brief introduction to weakly supervised learning. NSR, 5(1):44–53, 2018.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Guozhang Liu (3 papers)
  2. Ting Liu (329 papers)
  3. Mengke Yuan (6 papers)
  4. Tao Pang (14 papers)
  5. Guangxing Yang (1 paper)
  6. Hao Fu (82 papers)
  7. Tao Wang (700 papers)
  8. Tongkui Liao (2 papers)

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets