Learning RoI Transformer for Oriented Object Detection in Aerial Images

Abstract

Object detection in aerial images is an active yet challenging task in computer vision because of the birdview perspective, the highly complex backgrounds, and the variant appearances of objects. Especially when detecting densely packed objects in aerial images, methods relying on horizontal proposals for common object detection often introduce mismatches between the Region of Interests (RoIs) and objects. This leads to the common misalignment between the final object classification confidence and localization accuracy. Although rotated anchors have been used to tackle this problem, the design of them always multiplies the number of anchors and dramatically increases the computational complexity. In this paper, we propose a RoI Transformer to address these problems. More precisely, to improve the quality of region proposals, we first designed a Rotated RoI (RRoI) learner to transform a Horizontal Region of Interest (HRoI) into a Rotated Region of Interest (RRoI). Based on the RRoIs, we then proposed a Rotated Position Sensitive RoI Align (RPS-RoI-Align) module to extract rotation-invariant features from them for boosting subsequent classification and regression. Our RoI Transformer is with light weight and can be easily embedded into detectors for oriented object detection. A simple implementation of the RoI Transformer has achieved state-of-the-art performances on two common and challenging aerial datasets, i.e., DOTA and HRSC2016, with a neglectable reduction to detection speed. Our RoI Transformer exceeds the deformable Position Sensitive RoI pooling when oriented bounding-box annotations are available. Extensive experiments have also validated the flexibility and effectiveness of our RoI Transformer. The results demonstrate that it can be easily integrated with other detector architectures and significantly improve the performances.

Gliding Vertex on the Horizontal Bounding Box for Multi-Oriented Object Detection

“…As depicted in Tab. 1, using the same network on DOTA [25], the proposed method improves [5] by 4.49% and 4.75% mAP with and without FPN, respectively. The proposed method outperforms [5] by 1.2% mAP on HRSC2016 [38].…”

“…Without any extra network design such as cascade refinement and attention mechanism, the proposed method outperforms some stateof-the-art methods on both DOTA and HRSC2016 and is more efficient in runtime. Specifically, For the experiment on DOTA, the proposed method without FPN [3] achieves 73.39% mAP, outperforming the state-of-the-art method [5] by 5.65% mAP. FPN [3] that exploits better multi-scale features is also beneficial for the proposed method, boosting Quantitative comparison with other methods on DOTA.…”

“…Recent advances in multi-oriented object detection are mainly driven by adaption of classical object detection methods using rotated bounding boxes [4], [5] or quadrangles [6], [7], [8] to represent multi-oriented objects. Though these existing adaptions of horizontal object detection methods to multi-oriented object detection have achieved promising results, they still face some limitations.…”

Clustered Object Detection in Aerial Images

“…The work of [6] proposes a coupled region-based CNNs for aerial vehicle detection. The approach of [7] investigates the problem of misalignment between Region of Interests (RoI) and objects in aerial image detection, and introduces a ROI transformer to address this issue. The algorithm in [35] presents a scale adaptive proposal network for object detection in aerial images.…”

Adaptive Period Embedding for Representing Oriented Objects in Aerial Images

“…ICPR ODAI [32] and CVPR DOTA [33] competitions are organized based on this dataset. [5] presents a two-stage R-CNN method with RoI Transformer, which, in the first step, proposes horizontal bounding boxes. The first R-CNN outputs oriented bounding boxes, and the inputting of the second R-CNN are oriented bounding boxes.…”