English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Object detection has made tremendous strides in computer vision. Small objectdetection with appearance degradation is a prominent challenge, especially foraerial observations. To collect sufficient positive/negative samples forheuristic training, most object detectors preset region anchors in order tocalculate Intersection-over-Union (IoU) against the ground-truthed data. Inthis case, small objects are frequently abandoned or mislabeled. In this paper,we present an effective Dynamic Enhancement Anchor (DEA) network to construct anovel training sample generator. Different from the other state-of-the-arttechniques, the proposed network leverages a sample discriminator to realizeinteractive sample screening between an anchor-based unit and an anchor-freeunit to generate eligible samples. Besides, multi-task joint training with aconservative anchor-based inference scheme enhances the performance of theproposed model while reducing computational complexity. The proposed schemesupports both oriented and horizontal object detection tasks. Extensiveexperiments on two challenging aerial benchmarks (i.e., DOTA and HRSC2016)indicate that our method achieves state-of-the-art performance in accuracy withmoderate inference speed and computational overhead for training. On DOTA, ourDEA-Net which integrated with the baseline of RoI-Transformer surpasses theadvanced method by 0.40% mean-Average-Precision (mAP) for oriented objectdetection with a weaker backbone network (ResNet-101 vs ResNet-152) and 3.08%mean-Average-Precision (mAP) for horizontal object detection with the samebackbone. Besides, our DEA-Net which integrated with the baseline of ReDetachieves the state-of-the-art performance by 80.37%. On HRSC2016, it surpassesthe previous best model by 1.1% using only 3 horizontal anchors.

Anchor Retouching via Model Interaction for Robust Object Detection in Aerial Images