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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

One-stage detector basically formulates object detection as denseclassification and localization. The classification is usually optimized byFocal Loss and the box location is commonly learned under Dirac deltadistribution. A recent trend for one-stage detectors is to introduce anindividual prediction branch to estimate the quality of localization, where thepredicted quality facilitates the classification to improve detectionperformance. This paper delves into the representations of the above threefundamental elements: quality estimation, classification and localization. Twoproblems are discovered in existing practices, including (1) the inconsistentusage of the quality estimation and classification between training andinference and (2) the inflexible Dirac delta distribution for localization whenthere is ambiguity and uncertainty in complex scenes. To address the problems,we design new representations for these elements. Specifically, we merge thequality estimation into the class prediction vector to form a jointrepresentation of localization quality and classification, and use a vector torepresent arbitrary distribution of box locations. The improved representationseliminate the inconsistency risk and accurately depict the flexibledistribution in real data, but contain continuous labels, which is beyond thescope of Focal Loss. We then propose Generalized Focal Loss (GFL) thatgeneralizes Focal Loss from its discrete form to the continuous version forsuccessful optimization. On COCO test-dev, GFL achieves 45.0\% AP usingResNet-101 backbone, surpassing state-of-the-art SAPD (43.5\%) and ATSS(43.6\%) with higher or comparable inference speed, under the same backbone andtraining settings. Notably, our best model can achieve a single-modelsingle-scale AP of 48.2\%, at 10 FPS on a single 2080Ti GPU. Code and modelsare available at https://github.com/implus/GFocal.

Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection