Shulou(Shulou.com)06/03 Report--

Author: Amusi

Date:2019-11-26

Official account of Wechat: CVer

Link: read the preface of 8 latest target detection papers (EfficientDet/EdgeNet/ASFF/RoIMix/SCL/EFGRNet, etc.)

Computer vision paper express column has not updated paper express for some time, during this period of time, there are many papers, but there are not many more eye-catching papers. In order to facilitate your reading, I have sorted out the target detection (Object Detection) papers. It is also recommended that you pay attention to the express delivery of computer vision papers, you can learn the good content of CV more quickly!

The target detection papers shared in this article will be synchronously pushed to github. Welcome to star/fork (click to read the original text, or you can visit it directly):

Amusi/awesome-object-detection github.com

Note:

The target detection papers shared in this paper include not only target detection papers that refresh COCO mAP records, but also target detection papers in pursuit of mAP and FPS trade-off: November 2019.

[1] SCL: Towards AccurateDomain Adaptive Object Detection via Gradient Detach Based Stacked Complementary Losses

Time: 20191108

Author: CMU& Indian Institute of Technology

Link: https://arxiv.org/abs/1911.0255 9

Note: adaptive target detection network in SCL domain has better performance than MAF (ICCV'19) and Strong-Weak (CVPR'19).

[2] Localization-aware Channel Pruning for Object Detection

Time: 20191109

Author: Huazhong University of Science and Technology

Link: https://arxiv.org/abs/1911.0223 7

Note: the performance is better than that of DCP and ThiNet, and the parameters of SSD pruning can be 70%.

[3] RoIMix: Proposal-Fusion among Multiple Images for Underwater Object Detection

Time: 20191112

Author: Peking University & Pengcheng Laboratory & Tencent

Link: https://arxiv.org/abs/1911.0302 9

Note: underwater target detection, there is something.

[4] Model Adaption Object Detection System for Robot

Time: 20191113

Author: Xi'an Jiaotong University

Link: https://arxiv.org/abs/1911.0271 8

Note: to solve the problem that the robot keeps stable detection of the object when it moves close to the object. Detection and speed performance is better than YOLOv3!

[5] EdgeNet: Balancing Accuracy and Performance for Edge-based Convolutional Neural Network Object Detectors

Time: 20191117

Author: University of Cyprus

Link: https://arxiv.org/abs/1911.0609 1

Note: the speed and accuracy of EdgeNet are better than Tiny-YOLO V3 and DroNet, and the power consumption is only 4W! It can run in real time on raspberry pie and CPU.

[6] Enriched Feature Guided Refinement Network for Object Detection

Time: 2019 (ICCV 2019)

Author: Tianjin University & IIAI

Http://openaccess.thecvf.com/c ontent_ICCV_2019/papers/Nie_Enriched_Feature_Guided_Refinement_Network_for_Object_Detection_ICCV_2019_paper.pdf

Code: https://github.com/Ranchentx/EF GRNet

Note: EFGRNet is an improved Single-Stage detection network based on SSD, which can reach 46ms/39.0mAP (512x512) on COCO and is now open source!

[7] EfficientDet: Scalable and Efficient Object Detection

Time: 20191122

Author: Google brain (Quoc V. L. Boss)

Link: https://arxiv.org/abs/1911.0907 0

Code: open source soon

Note: BiFPN and EfficientDet are proposed in this paper, up to 51.0 mAP on COCO! It is the strongest target detection network without multi-scale testing at present.

[8] Learning Spatial Fusion for Single-Shot Object Detection

Time: 20191122

Author: Beijing University of Aeronautics and Astronautics

Link: https://arxiv.org/abs/1911.0951 6

Code: https://github.com/ruinmessi/AS FF

Note: YOLOv3+ASFF (Adaptive Spatial feature Fusion) combination, the performance is better than CornerNet and CenterNet, on COCO, 38.1mAP/60 FPS,43.9mAP/29FPS!

In order to facilitate downloading, I have packaged the above papers:

Baidu Cloud Link: https://pan.baidu.com/s/1muHIJr DKu-uQcagArfmOLA

Extraction code: yjqu

For more target detection papers, please see: https://github.com/amusi/awesom e-object-detection

If you CVer like this inventory, please give this article a like, if many people like, other CV direction of the large inventory series will also be launched as soon as possible!

Posted on 2019-11-26

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.