2022
DOI: 10.1609/aaai.v36i1.19975
|View full text |Cite
|
Sign up to set email alerts
|

Shape-Adaptive Selection and Measurement for Oriented Object Detection

Abstract: The development of detection methods for oriented object detection remains a challenging task. A considerable obstacle is the wide variation in the shape (e.g., aspect ratio) of objects. Sample selection in general object detection has been widely studied as it plays a crucial role in the performance of the detection method and has achieved great progress. However, existing sample selection strategies still overlook some issues: (1) most of them ignore the object shape information; (2) they do not make a poten… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1
1

Citation Types

0
17
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
4
4

Relationship

0
8

Authors

Journals

citations
Cited by 125 publications
(17 citation statements)
references
References 43 publications
0
17
0
Order By: Relevance
“…Compared with ATSS, our method achieves a 1.3% performance gain of the mAP. DAL [38] and SASM [39] are state-of-the-art label assignment methods for rotated object detection. Compared with DAL, our method achieves 0.8% performance gain of the mAP.…”
Section: Comparison With State-of-the-art Methodsmentioning
confidence: 99%
See 1 more Smart Citation
“…Compared with ATSS, our method achieves a 1.3% performance gain of the mAP. DAL [38] and SASM [39] are state-of-the-art label assignment methods for rotated object detection. Compared with DAL, our method achieves 0.8% performance gain of the mAP.…”
Section: Comparison With State-of-the-art Methodsmentioning
confidence: 99%
“…Ming et al [38] proposed dynamic anchor learning for arbitraryoriented object detection. In [39], an ATSS-based flexible shapeadaptive selection and shape-adaptive measurement strategy was proposed for oriented object detection. Different from the previous method ATSS, which only adopts the IoU to reflect the anchor quality, we propose the novel anchor quality function that takes into account both prior and prediction information of the anchor, which is more comprehensive.…”
Section: B Label Assignment In Object Detectionmentioning
confidence: 99%
“…CFC-Net [ 28 ] proposes a task-specific polarization function to solve the feature incompatibility problem of classification and regression tasks; ReDet [ 8 ] extracts rotation-invariant features by reconstructing the backbone network so that the feature maps are discriminative for objects with arbitrary orientations. SASM [ 29 ] proposes a shape-sensitive label assignment method that is suitable for high aspect ratio remote-sensing targets. RBox [ 30 ] improves the performance of the Transformer architecture in rotated object detection and solves the problem of redundant feature backgrounds.…”
Section: Methodsmentioning
confidence: 99%
“…This section compares the proposed YOLOOD with ten other existing arbitrary-oriented object detection methods: CSL [17], Gliding Vertex [42], Faster RCNN-O [9], R 3 Det[15], S 2 A-Net [28], KFIoU [51], SASM Reppoints [52], Oriented R-CNN [29], ReDet [53], ROI-Trans [27]. Table 2 summarizes the quantitative comparison results of the ten methods on the FFC dataset.…”
Section: Comparison With State Of the Artsmentioning
confidence: 99%
“…Good test results are also obtained for samples with varying sizes, angles, and dense objects. CSL [17] 71.42 39.50 Gliding Vertex [42] 73.04 36.20 Faster RCNN-O [9] 73.87 36.70 R 3 Det [15] 78.41 29.50 S 2 A-Net [28] 79.09 32.60 KFIoU [51] 82.52 29.30 SASM Reppoints [52] 82.82 34.60 Oriented R-CNN [29] 83.59 35.70 ReDet [53] 87.86 5.40 ROI-Trans [27] 90.92 29.30 YOLOOD(ours) 90.82 112.00…”
Section: Comparison With State Of the Artsmentioning
confidence: 99%