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

Knowledge distillation (KD) has shown potential for learning compact modelsin dense object detection. However, the commonly used softmax-baseddistillation ignores the absolute classification scores for individualcategories. Thus, the optimum of the distillation loss does not necessarilylead to the optimal student classification scores for dense object detectors.This cross-task protocol inconsistency is critical, especially for dense objectdetectors, since the foreground categories are extremely imbalanced. To addressthe issue of protocol differences between distillation and classification, wepropose a novel distillation method with cross-task consistent protocols,tailored for the dense object detection. For classification distillation, weaddress the cross-task protocol inconsistency problem by formulating theclassification logit maps in both teacher and student models as multiplebinary-classification maps and applying a binary-classification distillationloss to each map. For localization distillation, we design an IoU-basedLocalization Distillation Loss that is free from specific network structuresand can be compared with existing localization distillation losses. Ourproposed method is simple but effective, and experimental results demonstrateits superiority over existing methods. Code is available athttps://github.com/TinyTigerPan/BCKD.

Bridging Cross-task Protocol Inconsistency for Distillation in Dense Object Detection