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

The better accuracy and efficiency trade-off has been a challenging problemin object detection. In this work, we are dedicated to studying keyoptimizations and neural network architecture choices for object detection toimprove accuracy and efficiency. We investigate the applicability of theanchor-free strategy on lightweight object detection models. We enhance thebackbone structure and design the lightweight structure of the neck, whichimproves the feature extraction ability of the network. We improve labelassignment strategy and loss function to make training more stable andefficient. Through these optimizations, we create a new family of real-timeobject detectors, named PP-PicoDet, which achieves superior performance onobject detection for mobile devices. Our models achieve better trade-offsbetween accuracy and latency compared to other popular models. PicoDet-S withonly 0.99M parameters achieves 30.6% mAP, which is an absolute 4.8% improvementin mAP while reducing mobile CPU inference latency by 55% compared toYOLOX-Nano, and is an absolute 7.1% improvement in mAP compared to NanoDet. Itreaches 123 FPS (150 FPS using Paddle Lite) on mobile ARM CPU when the inputsize is 320. PicoDet-L with only 3.3M parameters achieves 40.9% mAP, which isan absolute 3.7% improvement in mAP and 44% faster than YOLOv5s. As shown inFigure 1, our models far outperform the state-of-the-art results forlightweight object detection. Code and pre-trained models are available athttps://github.com/PaddlePaddle/PaddleDetection.

PP-PicoDet: A Better Real-Time Object Detector on Mobile Devices