Abstract:Oriented object detection has been developed rapidly in the past few years, where rotation equivariance is crucial for detectors to predict rotated boxes. It is expected that the prediction can maintain the corresponding rotation when objects rotate, but severe mutation in angular prediction is sometimes observed when objects rotate near the boundary angle, which is well-known boundary discontinuity problem. The problem has been long believed to be caused by the sharp loss increase at the angular boundary, and widely used joint-optim IoU-like methods deal with this problem by loss-smoothing. However, we experimentally find that even state-of-the-art IoU-like methods actually fail to solve the problem. On further analysis, we find that the key to solution lies in encoding mode of the smoothing function rather than in joint or independent optimization. In existing IoU-like methods, the model essentially attempts to fit the angular relationship between box and object, where the break point at angular boundary makes the predictions highly <a class="link-external link-http" href="http://unstable.To" rel="external noopener nofollow">this http URL</a> deal with this issue, we propose a dual-optimization paradigm for angles. We decouple reversibility and joint-optim from single smoothing function into two distinct entities, which for the first time achieves the objectives of both correcting angular boundary and blending angle with other parameters.Extensive experiments on multiple datasets show that boundary discontinuity problem is well-addressed. Moreover, typical IoU-like methods are improved to the same level without obvious performance gap. The code is available at <a class="link-external link-https" href="https://github.com/hangxu-cv/cvpr24acm" rel="external noopener nofollow">this https URL</a>.

ABFL: Angular Boundary Discontinuity Free Loss for Arbitrary Oriented Object Detection in Aerial Images

ABFL: Angular Boundary Discontinuity Free Loss for Arbitrary Oriented Object Detection in Aerial Images

Rethinking Boundary Discontinuity Problem for Oriented Object Detection

ACE: Anchor-Free Corner Evolution for Real-Time Arbitrarily-Oriented Object Detection

Highly Efficient Anchor-Free Oriented Small Object Detection for Remote Sensing Images via Periodic Pseudo-Domain

Category-Aware Dynamic Label Assignment with High-Quality Oriented Proposal

Oriented RepPoints for Aerial Object Detection

Gaussian Focal Loss: Learning Distribution Polarized Angle Prediction for Rotated Object Detection in Aerial Images

High-Quality Angle Prediction for Oriented Object Detection in Remote Sensing Images.

Adaptive Period Embedding for Representing Oriented Objects in Aerial Images.

SAFDet: A Semi-Anchor-Free Detector for Effective Detection of Oriented Objects in Aerial Images

Improving Oriented Object Detection by Scene Classification and Task-Aligned Focal Loss

Multi-Oriented Object Detection in Aerial Images with Double Horizontal Rectangles

Building a Bridge of Bounding Box Regression Between Oriented and Horizontal Object Detection in Remote Sensing Images

Dynamic Anchor Learning for Arbitrary-Oriented Object Detection

AFDet: alignment and focusing for aerial object detection

Dense Label Encoding for Boundary Discontinuity Free Rotation Detection.

Theoretically Achieving Continuous Representation of Oriented Bounding Boxes

Fast arbitrary-oriented object detection for remote sensing images

Oriented Object Detection with Fine-Grained Enhancement and Angle Constraint

On the Arbitrary-Oriented Object Detection: Classification Based Approaches Revisited