当前位置:网站首页>Fast target detection (CS CV) based on grid multi-scale feature fusion

Fast target detection (CS CV) based on grid multi-scale feature fusion

2020-12-07 19:16:49 Ling Qian

Scale variation is a key problem in multiscale target detection . Early methods solved this problem by using image and feature pyramids , This produces suboptimal results under the constraints of computational burden and inherent network structure . The pioneering work also proposed multi-scale ( That is, multi-level 、 Multiple branches ) To make up for this problem , And has made gratifying progress . However , The existing fusion methods still have some limitations , If the feature scale is inconsistent 、 Ignore hierarchical semantic transformation 、 Coarse grain size, etc . In this work , We propose a new module ,Fluff block , In order to reduce the shortcomings of existing multiscale fusion methods and promote multiscale target detection . To be specific ,Fluff The multi-level and multi branch schemes with dilated convolution are used to realize fast 、 Effective and fine-grained feature fusion . Besides , We will Fluff And SSD Set into FluffNet, A powerful real-time single-stage multiscale target detection detector . stay MS-COCO and PASCAL-VOC The result of the experiment shows that ,flufffnet Remarkable efficiency at state-of-the-art accuracy . Besides , We also show how to embed it into other widely used detectors , Illustrates the Fluff The universality of blocks .

Original title :Fast Object Detection with Latticed Multi-Scale Feature Fusion

original text :Scale variance is one of the crucial challenges in multi-scale object detection. Early approaches address this problem by exploiting the image and feature pyramid, which raises suboptimal results with computation burden and constrains from inherent network structures. Pioneering works also propose multi-scale (i.e., multi-level and multi-branch) feature fusions to remedy the issue and have achieved encouraging progress. However, existing fusions still have certain limitations such as feature scale inconsistency, ignorance of level-wise semantic transformation, and coarse granularity. In this work, we present a novel module, the Fluff block, to alleviate drawbacks of current multi-scale fusion methods and facilitate multi-scale object detection. Specifically, Fluff leverages both multi-level and multi-branch schemes with dilated convolutions to have rapid, effective and finer-grained feature fusions. Furthermore, we integrate Fluff to SSD as FluffNet, a powerful real-time single-stage detector for multi-scale object detection. Empirical results on MS COCO and PASCAL VOC have demonstrated that FluffNet obtains remarkable efficiency with state-of-the-art accuracy. Additionally, we indicate the great generality of the Fluff block by showing how to embed it to other widely-used detectors as well.

Original author :Yue Shi, Bo Jiang, Zhengping Che, Jian Tang

Original address :https://arxiv.org/abs/2011.02780

Original statement , This article is authorized by the author + Community publication , Unauthorized , Shall not be reproduced .

If there is any infringement , Please contact the yunjia_community@tencent.com Delete .

版权声明
本文为[Ling Qian]所创,转载请带上原文链接,感谢
https://chowdera.com/2020/11/20201112182545431j.html