Polygon-to-Polygon Distance Loss for Rotated Object Detection

Yang, Yang; Chen, Jifeng; Zhong, Xiaopin; Deng, Yuanlong

doi:10.1609/aaai.v36i3.20214

Cited by 40 publications

(55 citation statements)

References 36 publications

(45 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, to avoid the boundary problem, we follow the GWD 11 to convert the oriented bounding box Bðx; y; w; h; θÞ into 2-D Gaussian distributions Nðμ; ΣÞ as follows:…”

Section: Modulated Kalman Intersection Over Union Lossmentioning

confidence: 99%

“…Some current methods 11 – 13 model the oriented bounding box into Gaussian distributions [see Eq. (1)] and then use the distance measure in statistics (such as Wassertein distance, 14 Kullback–Leibler divergence, 15 and Bhattacharyya distance 16 ) to calculate the similarity, effectively avoiding the boundary problems.…”

Section: Introductionmentioning

confidence: 99%

“…However, there is no clear correspondence between the calculated similarity and SkewIoU. On this basis, Yang et al 17 . used Kalman filter to mimic the computing mechanism of SkewIoU, but the inconsistency between loss and the evaluation metric still exists.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

MKIoU loss: toward accurate oriented object detection in aerial images

et al. 2023

J. Electron. Imag.

View full text Add to dashboard Cite

.Oriented bounding box regression is crucial for oriented object detection. However, regression-based methods often suffer from boundary problems and the inconsistency between loss and evaluation metrics. A modulated Kalman intersection over union (IoU) loss of approximate SkewIoU is proposed, named MKIoU. To avoid boundary problems, we convert the oriented bounding box to Gaussian distribution then use the Kalman filter to approximate the intersection area. However, there exists significant difference between the calculated and actual intersection areas. Thus, we propose a modulation factor to adjust the sensitivity of angle deviation and width-height offset to loss variation, making the loss more consistent with the evaluation metric. Furthermore, the Gaussian modeling method avoids the boundary problem but causes the angle confusion of square objects simultaneously. Thus, the Gaussian angle loss (GA loss) is presented to solve this problem by adding a corrected loss for square targets. The proposed GA loss can be easily extended to other Gaussian-based methods. Experiments on three publicly available aerial image datasets, DOTA, UCAS-AOD, and HRSC2016, show the effectiveness of the proposed method.

show abstract

“…First, to avoid the boundary problem, we follow the GWD 11 to convert the oriented bounding box Bðx; y; w; h; θÞ into 2-D Gaussian distributions Nðμ; ΣÞ as follows:…”

Section: Modulated Kalman Intersection Over Union Lossmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

MKIoU loss: toward accurate oriented object detection in aerial images

et al. 2023

J. Electron. Imag.

View full text Add to dashboard Cite

show abstract

“…SCRDet [2] combines SkewIoU and Smooth L1 loss to develop an IoU-Smooth L1 loss, which partly circumvents the need for differentiable SkewIoU loss. Polygon-to-Polygon distance loss [46] is derived from the area sum of triangles specified by the vertexes of one polygon and the edges of the other. KFIoU [47] achieves a trend-level alignment with SkewIoU by Gaussian modeling and Kalman filtering.…”

Section: Inconsistency Between Metric and Rotation Lossmentioning

confidence: 99%

Detecting Rotated Objects as Gaussian Distributions and Its 3-D Generalization

Yang

Gefan

Yang

et al. 2022

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

Existing detection methods commonly use a parameterized bounding box (BBox) to model and detect (horizontal) objects and an additional rotation angle parameter is used for rotated objects. We argue that such a mechanism has fundamental limitations in building an effective regression loss for rotation detection, especially for high-precision detection with high IoU (e.g. 0.75). Instead, we propose to model the rotated objects as Gaussian distributions. A direct advantage is that our new regression loss regarding the distance between two Gaussians e.g. Kullback-Leibler Divergence (KLD), can well align the actual detection performance metric, which is not well addressed in existing methods. Moreover, the two bottlenecks i.e. boundary discontinuity and square-like problem also disappear. We also propose an efficient Gaussian metric-based label assignment strategy to further boost the performance. Interestingly, by analyzing the BBox parameters' gradients under our Gaussian-based KLD loss, we show that these parameters are dynamically updated with interpretable physical meaning, which help explain the effectiveness of our approach, especially for high-precision detection. We extend our approach from 2-D to 3-D with a tailored algorithm design to handle the heading estimation, and experimental results on twelve public datasets (2-D/3-D, aerial/text/face images) with various base detectors show its superiority.

show abstract

“…There are so many researches are devoted to the object detection of remote sensing images, such as the ReDet [1], GWD [2], S2A-Net [3]. Most object detection methods involve classification, but the categories are limited to coarsegrained classifications, such as airplanes, ships and so on.…”

Section: Introductionmentioning

confidence: 99%

RAFF: remote sensing airplane fine grained classification with few shot learning

Wang

2023

Third International Conference on Computer Vision and Information Technology (CVIT 2022)

View full text Add to dashboard Cite

The fine-grained classification of remote sensing airplane images is a very meaningful work. Few of existing works have paid attention to the fine grained classification of remote sensing objects. The purpose of our research is to develop a better fine-grained classification performance of remote sensing airplane images. In this paper we propose a remote sensing airplane fine grained classification method with few shot learning. Few shot learning is used to alleviate the extreme imbalance distribution of the samples in different categories. We found two factors that affect the classification accuracy, the direction of the airplane and the background distribution. In order to increase the accuracy of classification and weaken the influence of the background, we propose an algorithm which use the symmetry of the image to predict the direction of the airplane, and add a Transpose encoder to alleviate the impact of the background distribution. The experimental results on Fair1m dataset proves the effectiveness of our method, which has obtained 5.73%up on accuracy performance.

show abstract

Polygon-to-Polygon Distance Loss for Rotated Object Detection

Cited by 40 publications

References 36 publications

MKIoU loss: toward accurate oriented object detection in aerial images

MKIoU loss: toward accurate oriented object detection in aerial images

Detecting Rotated Objects as Gaussian Distributions and Its 3-D Generalization

RAFF: remote sensing airplane fine grained classification with few shot learning

Contact Info

Product

Resources

About