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

Anchor-free detectors basically formulate object detection as denseclassification and regression. For popular anchor-free detectors, it is commonto introduce an individual prediction branch to estimate the quality oflocalization. The following inconsistencies are observed when we delve into thepractices of classification and quality estimation. Firstly, for some adjacentsamples which are assigned completely different labels, the trained model wouldproduce similar classification scores. This violates the training objective andleads to performance degradation. Secondly, it is found that detected boundingboxes with higher confidences contrarily have smaller overlaps with thecorresponding ground-truth. Accurately localized bounding boxes would besuppressed by less accurate ones in the Non-Maximum Suppression (NMS)procedure. To address the inconsistency problems, the Dynamic Smooth LabelAssignment (DSLA) method is proposed. Based on the concept of centernessoriginally developed in FCOS, a smooth assignment strategy is proposed. Thelabel is smoothed to a continuous value in [0, 1] to make a steady transitionbetween positive and negative samples. Intersection-of-Union (IoU) is predicteddynamically during training and is coupled with the smoothed label. The dynamicsmooth label is assigned to supervise the classification branch. Under suchsupervision, quality estimation branch is naturally merged into theclassification branch, which simplifies the architecture of anchor-freedetector. Comprehensive experiments are conducted on the MS COCO benchmark. Itis demonstrated that, DSLA can significantly boost the detection accuracy byalleviating the above inconsistencies for anchor-free detectors. Our codes arereleased at https://github.com/YonghaoHe/DSLA.

DSLA: Dynamic smooth label assignment for efficient anchor-free object detection