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 Tools annual

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 monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Scale-sensitive object detection remains a challenging task, where most ofthe existing methods could not learn it explicitly and are not robust to scalevariance. In addition, the most existing methods are less efficient duringtraining or slow during inference, which are not friendly to real-timeapplications. In this paper, we propose a practical object detection methodwith scale-sensitive network.Our method first predicts a global continuousscale ,which is shared by all position, for each convolution filter of eachnetwork stage. To effectively learn the scale, we average the spatial featuresand distill the scale from channels. For fast-deployment, we propose a scaledecomposition method that transfers the robust fractional scale intocombination of fixed integral scales for each convolution filter, whichexploits the dilated convolution. We demonstrate it on one-stage and two-stagealgorithms under different configurations. For practical applications, trainingof our method is of efficiency and simplicity which gets rid of complex datasampling or optimize strategy. During test-ing, the proposed method requires noextra operation and is very supportive of hardware acceleration like TensorRTand TVM. On the COCO test-dev, our model could achieve a 41.5 mAP on one-stagedetector and 42.1 mAP on two-stage detectors based on ResNet-101, outperformingbase-lines by 2.4 and 2.1 respectively without extra FLOPS.

POD: Practical Object Detection with Scale-Sensitive Network